diff --git a/src/calibre/ebooks/conversion/plugins/pdf_input.py b/src/calibre/ebooks/conversion/plugins/pdf_input.py
index 3bbe33207d..20cc20d687 100644
--- a/src/calibre/ebooks/conversion/plugins/pdf_input.py
+++ b/src/calibre/ebooks/conversion/plugins/pdf_input.py
@@ -67,5 +67,12 @@ class PDFInput(InputFormatPlugin):
         log.debug('Rendering manifest...')
         with open(u'metadata.opf', 'wb') as opffile:
             opf.render(opffile)
+        if os.path.exists(u'toc.ncx'):
+            ncxid = opf.manifest.id_for_path('toc.ncx')
+            if ncxid:
+                with open(u'metadata.opf', 'r+b') as f:
+                    raw = f.read().replace(b'<spine', b'<spine toc="%s"' % bytes(ncxid))
+                    f.seek(0)
+                    f.write(raw)
 
         return os.path.join(os.getcwdu(), u'metadata.opf')
diff --git a/src/calibre/ebooks/conversion/preprocess.py b/src/calibre/ebooks/conversion/preprocess.py
index f5660b95e8..bf181bb7c0 100644
--- a/src/calibre/ebooks/conversion/preprocess.py
+++ b/src/calibre/ebooks/conversion/preprocess.py
@@ -460,8 +460,6 @@ class HTMLPreProcessor(object):
                   # Center separator lines
                   (re.compile(u'<br>\s*(?P<break>([*#•✦=] *){3,})\s*<br>'), lambda match: '<p>\n<p style="text-align:center">' + match.group('break') + '</p>'),
 
-                  # Remove page links
-                  (re.compile(r'<a name=\d+></a>', re.IGNORECASE), lambda match: ''),
                   # Remove <hr> tags
                   (re.compile(r'<hr.*?>', re.IGNORECASE), lambda match: ''),
 
diff --git a/src/calibre/ebooks/pdf/pdftohtml.py b/src/calibre/ebooks/pdf/pdftohtml.py
index c9e3c32395..fe5bbde73a 100644
--- a/src/calibre/ebooks/pdf/pdftohtml.py
+++ b/src/calibre/ebooks/pdf/pdftohtml.py
@@ -85,10 +85,6 @@ def pdftohtml(output_dir, pdf_path, no_images, as_xml=False):
         logf.flush()
         logf.close()
         out = open(logf.name, 'rb').read().strip()
-        try:
-            os.remove(pdfsrc)
-        except:
-            pass
         if ret != 0:
             raise ConversionError(b'return code: %d\n%s' % (ret, out))
         if out:
@@ -106,7 +102,46 @@ def pdftohtml(output_dir, pdf_path, no_images, as_xml=False):
                 i.truncate()
                 # versions of pdftohtml >= 0.20 output self closing <br> tags, this
                 # breaks the pdf heuristics regexps, so replace them
-                i.write(raw.replace(b'<br/>', b'<br>'))
+                raw = raw.replace(b'<br/>', b'<br>')
+                raw = re.sub(br'<a\s+name=(\d+)', br'<a id="\1"', raw, flags=re.I)
+                i.write(raw)
+
+            cmd = [exe, b'-f', b'1', '-l', '1', b'-xml', b'-i', b'-enc', b'UTF-8', b'-noframes', b'-p', b'-nomerge',
+                    b'-nodrm', b'-q', b'-stdout', a(pdfsrc)]
+            p = popen(cmd, stdout=subprocess.PIPE)
+            raw = p.stdout.read().strip()
+            if p.wait() == 0 and raw:
+                parse_outline(raw, output_dir)
+
+            if isbsd:
+                cmd.remove(b'-nodrm')
+
+        try:
+            os.remove(pdfsrc)
+        except:
+            pass
+
+def parse_outline(raw, output_dir):
+    from lxml import etree
+    outline = etree.fromstring(raw).xpath('(//outline)[1]')
+    if outline:
+        from calibre.ebooks.oeb.polish.toc import TOC, create_ncx
+        outline = outline[0]
+        toc = TOC()
+
+        def process_node(node, toc):
+            for child in node.iterdescendants('*'):
+                if child.tag == 'outline':
+                    parent = toc.children[-1] if toc.children else toc
+                    process_node(child, parent)
+                else:
+                    page = child.get('page', '1')
+                    toc.add(child.text, 'index.html', page)
+        process_node(outline, toc)
+        root = create_ncx(toc, (lambda x:x), 'pdftohtml', 'en', 'pdftohtml')
+        with open(os.path.join(output_dir, 'toc.ncx'), 'wb') as f:
+            f.write(etree.tostring(root, pretty_print=True, with_tail=False, encoding='utf-8', xml_declaration=True))
+
 
 def flip_image(img, flip):
     from calibre.utils.magick import Image