diff --git a/src/calibre/ebooks/conversion/utils.py b/src/calibre/ebooks/conversion/utils.py
index 15522d25e6..d9e5246223 100644
--- a/src/calibre/ebooks/conversion/utils.py
+++ b/src/calibre/ebooks/conversion/utils.py
@@ -23,6 +23,7 @@ class HeuristicProcessor(object):
         self.min_chapters = 1
         self.chapters_no_title = 0
         self.chapters_with_title = 0
+        self.blanks_deleted = False
         self.linereg = re.compile('(?<=<p).*?(?=</p>)', re.IGNORECASE|re.DOTALL)
         self.blankreg = re.compile(r'\s*(?P<openline><p(?!\sid=\"softbreak\")[^>]*>)\s*(?P<closeline></p>)', re.IGNORECASE)
         self.multi_blank = re.compile(r'(\s*<p[^>]*>\s*</p>){2,}', re.IGNORECASE)
@@ -422,6 +423,7 @@ class HeuristicProcessor(object):
         # blank paragraphs then delete blank lines to clean up spacing
         if blanks_between_paragraphs and getattr(self.extra_opts, 'delete_blank_paragraphs', False):
             self.log.debug("deleting blank lines")
+            self.blanks_deleted = True
             html = self.multi_blank.sub('\n<p id="softbreak" style="margin-top:1.5em; margin-bottom:1.5em"> </p>', html)
             html = self.blankreg.sub('', html)
 
@@ -479,6 +481,9 @@ class HeuristicProcessor(object):
         if getattr(self.extra_opts, 'format_scene_breaks', False):
             # Center separator lines
             html = re.sub(u'<(?P<outer>p|div)[^>]*>\s*(<(?P<inner1>font|span|[ibu])[^>]*>)?\s*(<(?P<inner2>font|span|[ibu])[^>]*>)?\s*(<(?P<inner3>font|span|[ibu])[^>]*>)?\s*(?P<break>([*#•]+\s*)+)\s*(</(?P=inner3)>)?\s*(</(?P=inner2)>)?\s*(</(?P=inner1)>)?\s*</(?P=outer)>', '<p style="text-align:center; margin-top:1.25em; margin-bottom:1.25em">' + '\g<break>' + '</p>', html)
+            if not self.blanks_deleted:
+                html = self.multi_blank.sub('\n<p id="softbreak" style="margin-top:1.5em; margin-bottom:1.5em"> </p>', html)
+            html = re.sub('<p\s+id="softbreak"[^>]*>\s*</p>', '<div id="softbreak" style="margin-left: 45%; margin-right: 45%; margin-top:1.5em; margin-bottom:1.5em"><hr style="height: 3px; background:#505050" /></div>', html)
 
         if self.deleted_nbsps:
             # put back non-breaking spaces in empty paragraphs to preserve original formatting
diff --git a/src/calibre/manual/conversion.rst b/src/calibre/manual/conversion.rst
index 3383708b72..96a8e30e3c 100644
--- a/src/calibre/manual/conversion.rst
+++ b/src/calibre/manual/conversion.rst
@@ -530,17 +530,18 @@ more blank lines are a paragraph boundary::
 
 TXT input supports a number of options to differentiate how paragraphs are detected.
 
-    :guilabel:`Auto`
-        Analyzes the text file and attempts to determine how paragraphs are defined.
+    :guilabel:`Paragraph Style: Auto`
+        Analyzes the text file and attempts to automatically determine how paragraphs are defined.  This
+        option will generally work fine, if you achieve undesirable results try one of the manual options.
 
-    :guilabel:`Treat each line as a paragraph`
+    :guilabel:`Paragraph Style: Single`
         Assumes that every line is a paragraph::
 
             This is the first.
             This is the second.
             This is the third.
         
-    :guilabel:`Assume print formatting`
+    :guilabel:`Paragraph Style: Print`
         Assumes that every paragraph starts with an indent (either a tab or 2+ spaces). Paragraphs end when
         the next line that starts with an indent is reached::
 
@@ -551,11 +552,13 @@ TXT input supports a number of options to differentiate how paragraphs are detec
             This is the
             third.
 
-    :guilabel:`Unformatted`
+    :guilabel:`Paragraph Style: Unformatted`
         Assumes that the document has no formatting, but does use hard line breaks.  Punctuation
         and median line length are used to attempt to re-create paragraphs.
 
-    :guilabel:`Process using Textile`
+    :guilabel:`Formatting Style: Auto`
+
+    :guilabel:`Formatting Style: Heuristic`
 
     :guilabel:`Process using markdown`
         |app| also supports running TXT input though a transformation preprocessor known as markdown. Markdown