diff --git a/src/css_selectors/__init__.py b/src/css_selectors/__init__.py
new file mode 100644
index 0000000000..191e1dac64
--- /dev/null
+++ b/src/css_selectors/__init__.py
@@ -0,0 +1,9 @@
+#!/usr/bin/env python2
+# vim:fileencoding=utf-8
+from __future__ import (unicode_literals, division, absolute_import,
+                        print_function)
+
+__license__ = 'GPL v3'
+__copyright__ = '2015, Kovid Goyal <kovid at kovidgoyal.net>'
+
+
diff --git a/src/css_selectors/errors.py b/src/css_selectors/errors.py
new file mode 100644
index 0000000000..17f2d9095b
--- /dev/null
+++ b/src/css_selectors/errors.py
@@ -0,0 +1,19 @@
+#!/usr/bin/env python2
+# vim:fileencoding=utf-8
+from __future__ import (unicode_literals, division, absolute_import,
+                        print_function)
+
+__license__ = 'GPL v3'
+__copyright__ = '2015, Kovid Goyal <kovid at kovidgoyal.net>'
+
+class SelectorError(ValueError):
+
+    """Common parent for SelectorSyntaxError and ExpressionError"""
+
+class SelectorSyntaxError(SelectorError, SyntaxError):
+
+    """Parsing a selector that does not match the grammar."""
+
+class ExpressionError(SelectorError):
+
+    """Unknown or unsupported selector (eg. pseudo-class)."""
diff --git a/src/css_selectors/ordered_set.py b/src/css_selectors/ordered_set.py
new file mode 100644
index 0000000000..b2ab50caaa
--- /dev/null
+++ b/src/css_selectors/ordered_set.py
@@ -0,0 +1,139 @@
+#!/usr/bin/env python2
+# vim:fileencoding=utf-8
+from __future__ import (unicode_literals, division, absolute_import,
+                        print_function)
+
+__license__ = 'GPL v3'
+__copyright__ = '2015, Kovid Goyal <kovid at kovidgoyal.net>'
+
+import collections
+
+SLICE_ALL = slice(None)
+
+def is_iterable(obj):
+    """
+    Are we being asked to look up a list of things, instead of a single thing?
+    We check for the `__iter__` attribute so that this can cover types that
+    don't have to be known by this module, such as NumPy arrays.
+
+    Strings, however, should be considered as atomic values to look up, not
+    iterables.
+
+    We don't need to check for the Python 2 `unicode` type, because it doesn't
+    have an `__iter__` attribute anyway.
+    """
+    return hasattr(obj, '__iter__') and not isinstance(obj, str)
+
+
+class OrderedSet(collections.MutableSet):
+    """
+    An OrderedSet is a custom MutableSet that remembers its order, so that
+    every entry has an index that can be looked up.
+    """
+    def __init__(self, iterable=None):
+        if iterable is None:
+            self.items = []
+            self.map = {}
+        else:
+            for item in iterable:
+                idx = self.map.get(item)
+                if idx is None:
+                    self.map[item] = len(self.items)
+                    self.items.append(item)
+
+    def __len__(self):
+        return len(self.items)
+
+    def __getitem__(self, index):
+        """
+        Get the item at a given index.
+
+        If `index` is a slice, you will get back that slice of items. If it's
+        the slice [:], exactly the same object is returned. (If you want an
+        independent copy of an OrderedSet, use `OrderedSet.copy()`.)
+
+        If `index` is an iterable, you'll get the OrderedSet of items
+        corresponding to those indices. This is similar to NumPy's
+        "fancy indexing".
+        """
+        if index == SLICE_ALL:
+            return self
+        elif hasattr(index, '__index__') or isinstance(index, slice):
+            result = self.items[index]
+            if isinstance(result, list):
+                return OrderedSet(result)
+            else:
+                return result
+        elif is_iterable(index):
+            return OrderedSet([self.items[i] for i in index])
+        else:
+            raise TypeError("Don't know how to index an OrderedSet by %r" %
+                    index)
+
+    def copy(self):
+        return OrderedSet(self)
+
+    def __getstate__(self):
+        return tuple(self)
+
+    def __setstate__(self, state):
+        self.__init__(state)
+
+    def __contains__(self, key):
+        return key in self.map
+
+    def add(self, key):
+        """
+        Add `key` as an item to this OrderedSet, then return its index.
+
+        If `key` is already in the OrderedSet, return the index it already
+        had.
+        """
+        index = self.map.get(key)
+        if index is None:
+            self.map[key] = index = len(self.items)
+            self.items.append(key)
+        return index
+
+    def index(self, key):
+        """
+        Get the index of a given entry, raising an IndexError if it's not
+        present.
+
+        `key` can be an iterable of entries that is not a string, in which case
+        this returns a list of indices.
+        """
+        if is_iterable(key):
+            return [self.index(subkey) for subkey in key]
+        return self.map[key]
+
+    def discard(self, key):
+        index = self.map.get(key)
+        if index is not None:
+            self.items.pop(index)
+            for item in self.items[index:]:
+                self.map[item] -= 1
+            return True
+        return False
+
+    def __iter__(self):
+        return iter(self.items)
+
+    def __reversed__(self):
+        return reversed(self.items)
+
+    def __repr__(self):
+        if not self:
+            return '%s()' % (self.__class__.__name__,)
+        return '%s(%r)' % (self.__class__.__name__, list(self))
+
+    def __eq__(self, other):
+        if isinstance(other, OrderedSet):
+            return len(self) == len(other) and self.items == other.items
+        try:
+            other_as_set = set(other)
+        except TypeError:
+            # If `other` can't be converted into a set, it's not equal.
+            return False
+        else:
+            return set(self) == other_as_set
diff --git a/src/css_selectors/parse.py b/src/css_selectors/parse.py
new file mode 100644
index 0000000000..e12ae5f920
--- /dev/null
+++ b/src/css_selectors/parse.py
@@ -0,0 +1,758 @@
+#!/usr/bin/env python2
+# vim:fileencoding=utf-8
+"""
+    Tokenizer, parser and parsed objects for CSS selectors.
+
+    :copyright: (c) 2007-2012 Ian Bicking and contributors.
+                See AUTHORS for more details.
+    :license: BSD, see LICENSE for more details.
+
+"""
+
+import sys
+import re
+import operator
+import string
+
+from css_selectors.errors import SelectorSyntaxError
+
+if sys.version_info[0] < 3:
+    _unicode = unicode
+    _unichr = unichr
+else:
+    _unicode = str
+    _unichr = chr
+
+tab = string.maketrans(string.ascii_uppercase, string.ascii_lowercase)
+utab = {c:c+32 for c in range(ord('A'), ord('Z')+1)}
+
+def ascii_lower(string):
+    """Lower-case, but only in the ASCII range."""
+    return string.translate(utab if isinstance(string, _unicode) else tab)
+
+# Parsed objects
+
+class Selector(object):
+
+    """
+    Represents a parsed selector.
+    """
+
+    def __init__(self, tree, pseudo_element=None):
+        self.parsed_tree = tree
+        if pseudo_element is not None and not isinstance(
+                pseudo_element, FunctionalPseudoElement):
+            pseudo_element = ascii_lower(pseudo_element)
+        #: A :class:`FunctionalPseudoElement`,
+        #: or the identifier for the pseudo-element as a string,
+        #  or ``None``.
+        #:
+        #: +-------------------------+----------------+--------------------------------+
+        #: |                         | Selector       | Pseudo-element                 |
+        #: +=========================+================+================================+
+        #: | CSS3 syntax             | ``a::before``  | ``'before'``                   |
+        #: +-------------------------+----------------+--------------------------------+
+        #: | Older syntax            | ``a:before``   | ``'before'``                   |
+        #: +-------------------------+----------------+--------------------------------+
+        #: | From the Lists3_ draft, | ``li::marker`` | ``'marker'``                   |
+        #: | not in Selectors3       |                |                                |
+        #: +-------------------------+----------------+--------------------------------+
+        #: | Invalid pseudo-class    | ``li:marker``  | ``None``                       |
+        #: +-------------------------+----------------+--------------------------------+
+        #: | Functinal               | ``a::foo(2)``  | ``FunctionalPseudoElement(…)`` |
+        #: +-------------------------+----------------+--------------------------------+
+        #:
+        # : .. _Lists3: http://www.w3.org/TR/2011/WD-css3-lists-20110524/#marker-pseudoelement
+        self.pseudo_element = pseudo_element
+
+    def __repr__(self):
+        if isinstance(self.pseudo_element, FunctionalPseudoElement):
+            pseudo_element = repr(self.pseudo_element)
+        if self.pseudo_element:
+            pseudo_element = '::%s' % self.pseudo_element
+        else:
+            pseudo_element = ''
+        return '%s[%r%s]' % (
+            self.__class__.__name__, self.parsed_tree, pseudo_element)
+
+    def specificity(self):
+        """Return the specificity_ of this selector as a tuple of 3 integers.
+
+        .. _specificity: http://www.w3.org/TR/selectors/#specificity
+
+        """
+        a, b, c = self.parsed_tree.specificity()
+        if self.pseudo_element:
+            c += 1
+        return a, b, c
+
+
+class Class(object):
+
+    """
+    Represents selector.class_name
+    """
+    def __init__(self, selector, class_name):
+        self.selector = selector
+        self.class_name = class_name
+
+    def __repr__(self):
+        return '%s[%r.%s]' % (
+            self.__class__.__name__, self.selector, self.class_name)
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        b += 1
+        return a, b, c
+
+
+class FunctionalPseudoElement(object):
+
+    """
+    Represents selector::name(arguments)
+
+    .. attribute:: name
+
+        The name (identifier) of the pseudo-element, as a string.
+
+    .. attribute:: arguments
+
+        The arguments of the pseudo-element, as a list of tokens.
+
+        **Note:** tokens are not part of the public API,
+        and may change between cssselect versions.
+        Use at your own risks.
+
+    """
+    def __init__(self, name, arguments):
+        self.name = ascii_lower(name)
+        self.arguments = arguments
+
+    def __repr__(self):
+        return '%s[::%s(%r)]' % (
+            self.__class__.__name__, self.name,
+            [token.value for token in self.arguments])
+
+    def argument_types(self):
+        return [token.type for token in self.arguments]
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        b += 1
+        return a, b, c
+
+
+class Function(object):
+
+    """
+    Represents selector:name(expr)
+    """
+    def __init__(self, selector, name, arguments):
+        self.selector = selector
+        self.name = ascii_lower(name)
+        self.arguments = arguments
+
+    def __repr__(self):
+        return '%s[%r:%s(%r)]' % (
+            self.__class__.__name__, self.selector, self.name,
+            [token.value for token in self.arguments])
+
+    def argument_types(self):
+        return [token.type for token in self.arguments]
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        b += 1
+        return a, b, c
+
+
+class Pseudo(object):
+
+    """
+    Represents selector:ident
+    """
+    def __init__(self, selector, ident):
+        self.selector = selector
+        self.ident = ascii_lower(ident)
+
+    def __repr__(self):
+        return '%s[%r:%s]' % (
+            self.__class__.__name__, self.selector, self.ident)
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        b += 1
+        return a, b, c
+
+
+class Negation(object):
+
+    """
+    Represents selector:not(subselector)
+    """
+    def __init__(self, selector, subselector):
+        self.selector = selector
+        self.subselector = subselector
+
+    def __repr__(self):
+        return '%s[%r:not(%r)]' % (
+            self.__class__.__name__, self.selector, self.subselector)
+
+    def specificity(self):
+        a1, b1, c1 = self.selector.specificity()
+        a2, b2, c2 = self.subselector.specificity()
+        return a1 + a2, b1 + b2, c1 + c2
+
+
+class Attrib(object):
+
+    """
+    Represents selector[namespace|attrib operator value]
+    """
+    def __init__(self, selector, namespace, attrib, operator, value):
+        self.selector = selector
+        self.namespace = namespace
+        self.attrib = attrib
+        self.operator = operator
+        self.value = value
+
+    def __repr__(self):
+        if self.namespace:
+            attrib = '%s|%s' % (self.namespace, self.attrib)
+        else:
+            attrib = self.attrib
+        if self.operator == 'exists':
+            return '%s[%r[%s]]' % (
+                self.__class__.__name__, self.selector, attrib)
+        else:
+            return '%s[%r[%s %s %r]]' % (
+                self.__class__.__name__, self.selector, attrib,
+                self.operator, self.value)
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        b += 1
+        return a, b, c
+
+
+class Element(object):
+
+    """
+    Represents namespace|element
+
+    `None` is for the universal selector '*'
+
+    """
+    def __init__(self, namespace=None, element=None):
+        self.namespace = namespace
+        self.element = element
+
+    def __repr__(self):
+        element = self.element or '*'
+        if self.namespace:
+            element = '%s|%s' % (self.namespace, element)
+        return '%s[%s]' % (self.__class__.__name__, element)
+
+    def specificity(self):
+        if self.element:
+            return 0, 0, 1
+        else:
+            return 0, 0, 0
+
+
+class Hash(object):
+
+    """
+    Represents selector#id
+    """
+    def __init__(self, selector, id):
+        self.selector = selector
+        self.id = id
+
+    def __repr__(self):
+        return '%s[%r#%s]' % (
+            self.__class__.__name__, self.selector, self.id)
+
+    def specificity(self):
+        a, b, c = self.selector.specificity()
+        a += 1
+        return a, b, c
+
+
+class CombinedSelector(object):
+
+    def __init__(self, selector, combinator, subselector):
+        assert selector is not None
+        self.selector = selector
+        self.combinator = combinator
+        self.subselector = subselector
+
+    def __repr__(self):
+        if self.combinator == ' ':
+            comb = '<followed>'
+        else:
+            comb = self.combinator
+        return '%s[%r %s %r]' % (
+            self.__class__.__name__, self.selector, comb, self.subselector)
+
+    def specificity(self):
+        a1, b1, c1 = self.selector.specificity()
+        a2, b2, c2 = self.subselector.specificity()
+        return a1 + a2, b1 + b2, c1 + c2
+
+
+# Parser
+
+# foo
+_el_re = re.compile(r'^[ \t\r\n\f]*([a-zA-Z]+)[ \t\r\n\f]*$')
+
+# foo#bar or #bar
+_id_re = re.compile(r'^[ \t\r\n\f]*([a-zA-Z]*)#([a-zA-Z0-9_-]+)[ \t\r\n\f]*$')
+
+# foo.bar or .bar
+_class_re = re.compile(
+    r'^[ \t\r\n\f]*([a-zA-Z]*)\.([a-zA-Z][a-zA-Z0-9_-]*)[ \t\r\n\f]*$')
+
+
+def parse(css):
+    """Parse a CSS *group of selectors*.
+
+    :param css:
+        A *group of selectors* as an Unicode string.
+    :raises:
+        :class:`SelectorSyntaxError` on invalid selectors.
+    :returns:
+        A list of parsed :class:`Selector` objects, one for each
+        selector in the comma-separated group.
+
+    """
+    # Fast path for simple cases
+    match = _el_re.match(css)
+    if match:
+        return [Selector(Element(element=match.group(1)))]
+    match = _id_re.match(css)
+    if match is not None:
+        return [Selector(Hash(Element(element=match.group(1) or None),
+                              match.group(2)))]
+    match = _class_re.match(css)
+    if match is not None:
+        return [Selector(Class(Element(element=match.group(1) or None),
+                               match.group(2)))]
+
+    stream = TokenStream(tokenize(css))
+    stream.source = css
+    return list(parse_selector_group(stream))
+#    except SelectorSyntaxError:
+#        e = sys.exc_info()[1]
+#        message = "%s at %s -> %r" % (
+#            e, stream.used, stream.peek())
+#        e.msg = message
+#        if sys.version_info < (2,6):
+#            e.message = message
+#        e.args = tuple([message])
+#        raise
+
+
+def parse_selector_group(stream):
+    stream.skip_whitespace()
+    while 1:
+        yield Selector(*parse_selector(stream))
+        if stream.peek() == ('DELIM', ','):
+            stream.next()
+            stream.skip_whitespace()
+        else:
+            break
+
+def parse_selector(stream):
+    result, pseudo_element = parse_simple_selector(stream)
+    while 1:
+        stream.skip_whitespace()
+        peek = stream.peek()
+        if peek in (('EOF', None), ('DELIM', ',')):
+            break
+        if pseudo_element:
+            raise SelectorSyntaxError(
+                'Got pseudo-element ::%s not at the end of a selector'
+                % pseudo_element)
+        if peek.is_delim('+', '>', '~'):
+            # A combinator
+            combinator = stream.next().value
+            stream.skip_whitespace()
+        else:
+            # By exclusion, the last parse_simple_selector() ended
+            # at peek == ' '
+            combinator = ' '
+        next_selector, pseudo_element = parse_simple_selector(stream)
+        result = CombinedSelector(result, combinator, next_selector)
+    return result, pseudo_element
+
+
+def parse_simple_selector(stream, inside_negation=False):
+    stream.skip_whitespace()
+    selector_start = len(stream.used)
+    peek = stream.peek()
+    if peek.type == 'IDENT' or peek == ('DELIM', '*'):
+        if peek.type == 'IDENT':
+            namespace = stream.next().value
+        else:
+            stream.next()
+            namespace = None
+        if stream.peek() == ('DELIM', '|'):
+            stream.next()
+            element = stream.next_ident_or_star()
+        else:
+            element = namespace
+            namespace = None
+    else:
+        element = namespace = None
+    result = Element(namespace, element)
+    pseudo_element = None
+    while 1:
+        peek = stream.peek()
+        if peek.type in ('S', 'EOF') or peek.is_delim(',', '+', '>', '~') or (
+                inside_negation and peek == ('DELIM', ')')):
+            break
+        if pseudo_element:
+            raise SelectorSyntaxError(
+                'Got pseudo-element ::%s not at the end of a selector'
+                % pseudo_element)
+        if peek.type == 'HASH':
+            result = Hash(result, stream.next().value)
+        elif peek == ('DELIM', '.'):
+            stream.next()
+            result = Class(result, stream.next_ident())
+        elif peek == ('DELIM', '['):
+            stream.next()
+            result = parse_attrib(result, stream)
+        elif peek == ('DELIM', ':'):
+            stream.next()
+            if stream.peek() == ('DELIM', ':'):
+                stream.next()
+                pseudo_element = stream.next_ident()
+                if stream.peek() == ('DELIM', '('):
+                    stream.next()
+                    pseudo_element = FunctionalPseudoElement(
+                        pseudo_element, parse_arguments(stream))
+                continue
+            ident = stream.next_ident()
+            if ident.lower() in ('first-line', 'first-letter',
+                                 'before', 'after'):
+                # Special case: CSS 2.1 pseudo-elements can have a single ':'
+                # Any new pseudo-element must have two.
+                pseudo_element = _unicode(ident)
+                continue
+            if stream.peek() != ('DELIM', '('):
+                result = Pseudo(result, ident)
+                continue
+            stream.next()
+            stream.skip_whitespace()
+            if ident.lower() == 'not':
+                if inside_negation:
+                    raise SelectorSyntaxError('Got nested :not()')
+                argument, argument_pseudo_element = parse_simple_selector(
+                    stream, inside_negation=True)
+                next = stream.next()
+                if argument_pseudo_element:
+                    raise SelectorSyntaxError(
+                        'Got pseudo-element ::%s inside :not() at %s'
+                        % (argument_pseudo_element, next.pos))
+                if next != ('DELIM', ')'):
+                    raise SelectorSyntaxError("Expected ')', got %s" % (next,))
+                result = Negation(result, argument)
+            else:
+                result = Function(result, ident, parse_arguments(stream))
+        else:
+            raise SelectorSyntaxError(
+                "Expected selector, got %s" % (peek,))
+    if len(stream.used) == selector_start:
+        raise SelectorSyntaxError(
+            "Expected selector, got %s" % (stream.peek(),))
+    return result, pseudo_element
+
+
+def parse_arguments(stream):
+    arguments = []
+    while 1:
+        stream.skip_whitespace()
+        next = stream.next()
+        if next.type in ('IDENT', 'STRING', 'NUMBER') or next in [
+                ('DELIM', '+'), ('DELIM', '-')]:
+            arguments.append(next)
+        elif next == ('DELIM', ')'):
+            return arguments
+        else:
+            raise SelectorSyntaxError(
+                "Expected an argument, got %s" % (next,))
+
+
+def parse_attrib(selector, stream):
+    stream.skip_whitespace()
+    attrib = stream.next_ident_or_star()
+    if attrib is None and stream.peek() != ('DELIM', '|'):
+        raise SelectorSyntaxError(
+            "Expected '|', got %s" % (stream.peek(),))
+    if stream.peek() == ('DELIM', '|'):
+        stream.next()
+        if stream.peek() == ('DELIM', '='):
+            namespace = None
+            stream.next()
+            op = '|='
+        else:
+            namespace = attrib
+            attrib = stream.next_ident()
+            op = None
+    else:
+        namespace = op = None
+    if op is None:
+        stream.skip_whitespace()
+        next = stream.next()
+        if next == ('DELIM', ']'):
+            return Attrib(selector, namespace, attrib, 'exists', None)
+        elif next == ('DELIM', '='):
+            op = '='
+        elif next.is_delim('^', '$', '*', '~', '|', '!') and (
+                stream.peek() == ('DELIM', '=')):
+            op = next.value + '='
+            stream.next()
+        else:
+            raise SelectorSyntaxError(
+                "Operator expected, got %s" % (next,))
+    stream.skip_whitespace()
+    value = stream.next()
+    if value.type not in ('IDENT', 'STRING'):
+        raise SelectorSyntaxError(
+            "Expected string or ident, got %s" % (value,))
+    stream.skip_whitespace()
+    next = stream.next()
+    if next != ('DELIM', ']'):
+        raise SelectorSyntaxError(
+            "Expected ']', got %s" % (next,))
+    return Attrib(selector, namespace, attrib, op, value.value)
+
+
+def parse_series(tokens):
+    """
+    Parses the arguments for :nth-child() and friends.
+
+    :raises: A list of tokens
+    :returns: :``(a, b)``
+
+    """
+    for token in tokens:
+        if token.type == 'STRING':
+            raise ValueError('String tokens not allowed in series.')
+    s = ''.join(token.value for token in tokens).strip()
+    if s == 'odd':
+        return (2, 1)
+    elif s == 'even':
+        return (2, 0)
+    elif s == 'n':
+        return (1, 0)
+    if 'n' not in s:
+        # Just b
+        return (0, int(s))
+    a, b = s.split('n', 1)
+    if not a:
+        a = 1
+    elif a == '-' or a == '+':
+        a = int(a+'1')
+    else:
+        a = int(a)
+    if not b:
+        b = 0
+    else:
+        b = int(b)
+    return (a, b)
+
+
+# Token objects
+
+class Token(tuple):
+
+    def __new__(cls, type_, value, pos):
+        obj = tuple.__new__(cls, (type_, value))
+        obj.pos = pos
+        return obj
+
+    def __repr__(self):
+        return "<%s '%s' at %i>" % (self.type, self.value, self.pos)
+
+    def is_delim(self, *values):
+        return self.type == 'DELIM' and self.value in values
+
+    type = property(operator.itemgetter(0))
+    value = property(operator.itemgetter(1))
+
+
+class EOFToken(Token):
+
+    def __new__(cls, pos):
+        return Token.__new__(cls, 'EOF', None, pos)
+
+    def __repr__(self):
+        return '<%s at %i>' % (self.type, self.pos)
+
+
+# Tokenizer
+
+
+class TokenMacros:
+    unicode_escape = r'\\([0-9a-f]{1,6})(?:\r\n|[ \n\r\t\f])?'
+    escape = unicode_escape + r'|\\[^\n\r\f0-9a-f]'
+    string_escape = r'\\(?:\n|\r\n|\r|\f)|' + escape
+    nonascii = r'[^\0-\177]'
+    nmchar = '[_a-z0-9-]|%s|%s' % (escape, nonascii)
+    nmstart = '[_a-z]|%s|%s' % (escape, nonascii)
+
+def _compile(pattern):
+    return re.compile(pattern % vars(TokenMacros), re.IGNORECASE).match
+
+_match_whitespace = _compile(r'[ \t\r\n\f]+')
+_match_number = _compile('[+-]?(?:[0-9]*\.[0-9]+|[0-9]+)')
+_match_hash = _compile('#(?:%(nmchar)s)+')
+_match_ident = _compile('-?(?:%(nmstart)s)(?:%(nmchar)s)*')
+_match_string_by_quote = {
+    "'": _compile(r"([^\n\r\f\\']|%(string_escape)s)*"),
+    '"': _compile(r'([^\n\r\f\\"]|%(string_escape)s)*'),
+}
+
+_sub_simple_escape = re.compile(r'\\(.)').sub
+_sub_unicode_escape = re.compile(TokenMacros.unicode_escape, re.I).sub
+_sub_newline_escape =re.compile(r'\\(?:\n|\r\n|\r|\f)').sub
+
+# Same as r'\1', but faster on CPython
+if hasattr(operator, 'methodcaller'):
+    # Python 2.6+
+    _replace_simple = operator.methodcaller('group', 1)
+else:
+    def _replace_simple(match):
+        return match.group(1)
+
+def _replace_unicode(match):
+    codepoint = int(match.group(1), 16)
+    if codepoint > sys.maxunicode:
+        codepoint = 0xFFFD
+    return _unichr(codepoint)
+
+
+def unescape_ident(value):
+    value = _sub_unicode_escape(_replace_unicode, value)
+    value = _sub_simple_escape(_replace_simple, value)
+    return value
+
+
+def tokenize(s):
+    pos = 0
+    len_s = len(s)
+    while pos < len_s:
+        match = _match_whitespace(s, pos=pos)
+        if match:
+            yield Token('S', ' ', pos)
+            pos = match.end()
+            continue
+
+        match = _match_ident(s, pos=pos)
+        if match:
+            value = _sub_simple_escape(_replace_simple,
+                    _sub_unicode_escape(_replace_unicode, match.group()))
+            yield Token('IDENT', value, pos)
+            pos = match.end()
+            continue
+
+        match = _match_hash(s, pos=pos)
+        if match:
+            value = _sub_simple_escape(_replace_simple,
+                    _sub_unicode_escape(_replace_unicode, match.group()[1:]))
+            yield Token('HASH', value, pos)
+            pos = match.end()
+            continue
+
+        quote = s[pos]
+        if quote in _match_string_by_quote:
+            match = _match_string_by_quote[quote](s, pos=pos + 1)
+            assert match, 'Should have found at least an empty match'
+            end_pos = match.end()
+            if end_pos == len_s:
+                raise SelectorSyntaxError('Unclosed string at %s' % pos)
+            if s[end_pos] != quote:
+                raise SelectorSyntaxError('Invalid string at %s' % pos)
+            value = _sub_simple_escape(_replace_simple,
+                    _sub_unicode_escape(_replace_unicode,
+                    _sub_newline_escape('', match.group())))
+            yield Token('STRING', value, pos)
+            pos = end_pos + 1
+            continue
+
+        match = _match_number(s, pos=pos)
+        if match:
+            value = match.group()
+            yield Token('NUMBER', value, pos)
+            pos = match.end()
+            continue
+
+        pos2 = pos + 2
+        if s[pos:pos2] == '/*':
+            pos = s.find('*/', pos2)
+            if pos == -1:
+                pos = len_s
+            else:
+                pos += 2
+            continue
+
+        yield Token('DELIM', s[pos], pos)
+        pos += 1
+
+    assert pos == len_s
+    yield EOFToken(pos)
+
+
+class TokenStream(object):
+
+    def __init__(self, tokens, source=None):
+        self.used = []
+        self.tokens = iter(tokens)
+        self.source = source
+        self.peeked = None
+        self._peeking = False
+        try:
+            self.next_token = self.tokens.next
+        except AttributeError:
+            # Python 3
+            self.next_token = self.tokens.__next__
+
+    def next(self):
+        if self._peeking:
+            self._peeking = False
+            self.used.append(self.peeked)
+            return self.peeked
+        else:
+            next = self.next_token()
+            self.used.append(next)
+            return next
+
+    def peek(self):
+        if not self._peeking:
+            self.peeked = self.next_token()
+            self._peeking = True
+        return self.peeked
+
+    def next_ident(self):
+        next = self.next()
+        if next.type != 'IDENT':
+            raise SelectorSyntaxError('Expected ident, got %s' % (next,))
+        return next.value
+
+    def next_ident_or_star(self):
+        next = self.next()
+        if next.type == 'IDENT':
+            return next.value
+        elif next == ('DELIM', '*'):
+            return None
+        else:
+            raise SelectorSyntaxError(
+                "Expected ident or '*', got %s" % (next,))
+
+    def skip_whitespace(self):
+        peek = self.peek()
+        if peek.type == 'S':
+            self.next()
diff --git a/src/css_selectors/select.py b/src/css_selectors/select.py
new file mode 100644
index 0000000000..77663364e8
--- /dev/null
+++ b/src/css_selectors/select.py
@@ -0,0 +1,213 @@
+#!/usr/bin/env python2
+# vim:fileencoding=utf-8
+from __future__ import (unicode_literals, division, absolute_import,
+                        print_function)
+
+__license__ = 'GPL v3'
+__copyright__ = '2015, Kovid Goyal <kovid at kovidgoyal.net>'
+
+from collections import OrderedDict, defaultdict
+from functools import wraps
+
+from lxml import etree
+
+from css_selectors.errors import ExpressionError
+from css_selectors.parse import parse, ascii_lower, Element
+from css_selectors.ordered_set import OrderedSet
+
+PARSE_CACHE_SIZE = 200
+parse_cache = OrderedDict()
+XPATH_CACHE_SIZE = 30
+xpath_cache = OrderedDict()
+
+def get_parsed_selector(raw):
+    try:
+        return parse_cache[raw]
+    except KeyError:
+        parse_cache[raw] = ans = parse(raw)
+        if len(parse_cache) > PARSE_CACHE_SIZE:
+            parse_cache.pop(next(iter(parse_cache)))
+        return ans
+
+def get_compiled_xpath(expr):
+    try:
+        return xpath_cache[expr]
+    except KeyError:
+        xpath_cache[expr] = ans = etree.XPath(expr)
+        if len(xpath_cache) > XPATH_CACHE_SIZE:
+            xpath_cache.pop(next(iter(xpath_cache)))
+        return ans
+
+class AlwaysIn(object):
+
+    def __contains__(self, x):
+        return True
+always_in = AlwaysIn()
+
+def trace_wrapper(func):
+    @wraps(func)
+    def trace(*args, **kwargs):
+        targs = args[1:] if args and isinstance(args[0], Select) else args
+        print ('Called:', func.__name__, 'with args:', targs, kwargs or '')
+        return func(*args, **kwargs)
+    return trace
+
+class Select(object):
+
+    combinator_mapping = {
+        ' ': 'descendant',
+        '>': 'child',
+        '+': 'direct_adjacent',
+        '~': 'indirect_adjacent',
+    }
+
+    attribute_operator_mapping = {
+        'exists': 'exists',
+        '=': 'equals',
+        '~=': 'includes',
+        '|=': 'dashmatch',
+        '^=': 'prefixmatch',
+        '$=': 'suffixmatch',
+        '*=': 'substringmatch',
+        '!=': 'different',  # Not in Level 3 but I like it ;)
+    }
+
+    def __init__(self, root, dispatch_map=None, trace=False):
+        self.root = root
+        self.dispatch_map = dispatch_map or default_dispatch_map
+        self.invalidate_caches()
+        if trace:
+            self.dispatch_map = {k:trace_wrapper(v) for k, v in self.dispatch_map.iteritems()}
+
+    def invalidate_caches(self):
+        self._element_map = None
+        self._id_map = None
+        self._class_map = None
+
+    def __call__(self, selector):
+        for selector in get_parsed_selector(selector):
+            parsed_selector = selector.parsed_tree
+            for item in self.iterparsedselector(parsed_selector):
+                yield item
+
+    def iterparsedselector(self, parsed_selector):
+        type_name = type(parsed_selector).__name__
+        try:
+            func = self.dispatch_map[ascii_lower(type_name)]
+        except KeyError:
+            raise ExpressionError('%s is not supported' % type_name)
+        for item in func(self, parsed_selector):
+            yield item
+
+    @property
+    def element_map(self):
+        if self._element_map is None:
+            self._element_map = em = defaultdict(OrderedSet)
+            map_tag_name = ascii_lower
+            if '{' in self.root.tag:
+                def map_tag_name(x):
+                    return ascii_lower(x.rpartition('}')[2])
+
+            for tag in root.iter('*'):
+                em[map_tag_name(tag.tag)].add(tag)
+        return self._element_map
+
+    @property
+    def id_map(self):
+        if self._id_map is None:
+            self._id_map = im = defaultdict(OrderedSet)
+            lower = ascii_lower
+            for elem in get_compiled_xpath('//*[@id]')(self.root):
+                im[lower(elem.get('id'))].add(elem)
+        return self._id_map
+
+    @property
+    def class_map(self):
+        if self._class_map is None:
+            self._class_map = cm = defaultdict(OrderedSet)
+            lower = ascii_lower
+            for elem in get_compiled_xpath('//*[@class]')(self.root):
+                for cls in elem.get('class').split():
+                    cm[lower(cls)].add(elem)
+        return self._class_map
+
+# Combinators {{{
+
+def select_combinedselector(cache, combined):
+    """Translate a combined selector."""
+    combinator = cache.combinator_mapping[combined.combinator]
+    # Fast path for when the sub-selector is all elements
+    right = None if isinstance(combined.subselector, Element) and (
+        combined.subselector.element or '*') == '*' else cache.iterparsedselector(combined.subselector)
+    for item in cache.dispatch_map[combinator](cache.iterparsedselector(combined.selector), right):
+        yield item
+
+def select_descendant(left, right):
+    """right is a child, grand-child or further descendant of left"""
+    right = always_in if right is None else frozenset(right)
+    for ancestor in left:
+        for descendant in ancestor.iterdescendants('*'):
+            if descendant in right:
+                yield descendant
+
+def select_child(left, right):
+    """right is an immediate child of left"""
+    right = always_in if right is None else frozenset(right)
+    for parent in left:
+        for child in parent.iterchildren('*'):
+            if child in right:
+                yield child
+
+def select_direct_adjacent(left, right):
+    """right is a sibling immediately after left"""
+    right = always_in if right is None else frozenset(right)
+    for parent in left:
+        for sibling in parent.itersiblings('*'):
+            if sibling in right:
+                yield sibling
+            break
+
+def select_indirect_adjacent(left, right):
+    """right is a sibling after left, immediately or not"""
+    right = always_in if right is None else frozenset(right)
+    for parent in left:
+        for sibling in parent.itersiblings('*'):
+            if sibling in right:
+                yield sibling
+# }}}
+
+def select_element(cache, selector):
+    """A type or universal selector."""
+    element = selector.element
+    if not element or element == '*':
+        for elem in cache.root.iter('*'):
+            yield elem
+    else:
+        for elem in cache.element_map[ascii_lower(element)]:
+            yield elem
+
+def select_hash(cache, selector):
+    'An id selector'
+    items = cache.id_map[ascii_lower(selector.id)]
+    if len(items) > 1:
+        for elem in cache.iterparsedselector(selector.selector):
+            if elem in items:
+                yield elem
+    elif items:
+        yield items[0]
+
+def select_class(cache, selector):
+    'A class selector'
+    items = cache.class_map[ascii_lower(selector.class_name)]
+    if items:
+        for elem in cache.iterparsedselector(selector.selector):
+            if elem in items:
+                yield elem
+
+default_dispatch_map = {name.partition('_')[2]:obj for name, obj in globals().items() if name.startswith('select_') and callable(obj)}
+
+if __name__ == '__main__':
+    from pprint import pprint
+    root = etree.fromstring('<body xmlns="xxx"><p id="p" class="one two"><a id="a"/></p></body>')
+    select = Select(root, trace=True)
+    pprint(list(select('p#p.one.two')))