diff --git a/recipes/people_us_mashup.recipe b/recipes/people_us_mashup.recipe
index 28c76d820c..5d820bacc0 100644
--- a/recipes/people_us_mashup.recipe
+++ b/recipes/people_us_mashup.recipe
@@ -18,6 +18,7 @@ class PeopleMag(BasicNewsRecipe):
 
     no_stylesheets = True
     auto_cleanup = True
+    auto_cleanup_keep = '//div[@id="article-image"]'
 
 
     feeds = [
diff --git a/src/calibre/ebooks/readability/readability.py b/src/calibre/ebooks/readability/readability.py
index 8d4a23b338..028a4d6ede 100644
--- a/src/calibre/ebooks/readability/readability.py
+++ b/src/calibre/ebooks/readability/readability.py
@@ -1,3 +1,8 @@
+#!/usr/bin/env python
+# vim:fileencoding=UTF-8:ts=4:sw=4:sta:et:sts=4:ai
+from __future__ import (unicode_literals, division, absolute_import,
+                        print_function)
+
 import re, sys
 from collections import defaultdict
 
@@ -72,10 +77,15 @@ def __init__(self, input, log, **options):
             self.options[k] = v
         self.html = None
         self.log = log
+        self.keep_elements = set()
 
     def _html(self, force=False):
         if force or self.html is None:
             self.html = self._parse(self.input)
+            path = self.options['keep_elements']
+            if path is not None:
+                self.keep_elements = set(self.html.xpath(path))
+
         return self.html
 
     def _parse(self, input):
@@ -152,8 +162,9 @@ def get_article(self, candidates, best_candidate):
             append = False
             if sibling is best_elem:
                 append = True
-            sibling_key = sibling #HashableElement(sibling)
-            if sibling_key in candidates and candidates[sibling_key]['content_score'] >= sibling_score_threshold:
+            if sibling in candidates and candidates[sibling]['content_score'] >= sibling_score_threshold:
+                append = True
+            if sibling in self.keep_elements:
                 append = True
 
             if sibling.tag == "p":
@@ -283,6 +294,8 @@ def debug(self, *a):
 
     def remove_unlikely_candidates(self):
         for elem in self.html.iter():
+            if elem in self.keep_elements:
+                continue
             s = "%s %s" % (elem.get('class', ''), elem.get('id', ''))
             #self.debug(s)
             if REGEXES['unlikelyCandidatesRe'].search(s) and (not REGEXES['okMaybeItsACandidateRe'].search(s)) and elem.tag != 'body':
@@ -337,7 +350,7 @@ def sanitize(self, node, candidates):
         allowed = {}
         # Conditionally clean <table>s, <ul>s, and <div>s
         for el in self.reverse_tags(node, "table", "ul", "div"):
-            if el in allowed:
+            if el in allowed or el in self.keep_elements:
                 continue
             weight = self.class_weight(el)
             if el in candidates:
@@ -450,46 +463,17 @@ def sanitize(self, node, candidates):
                     #self.debug("pname %s pweight %.3f" %(pname, pweight))
                     el.drop_tree()
 
-        for el in ([node] + [n for n in node.iter()]):
-            if not (self.options['attributes']):
-                #el.attrib = {} #FIXME:Checkout the effects of disabling this
-                pass
-
         return clean_attributes(tounicode(node))
 
-
-class HashableElement():
-    def __init__(self, node):
-        self.node = node
-        self._path = None
-
-    def _get_path(self):
-        if self._path is None:
-            reverse_path = []
-            node = self.node
-            while node is not None:
-                node_id = (node.tag, tuple(node.attrib.items()), node.text)
-                reverse_path.append(node_id)
-                node = node.getparent()
-            self._path = tuple(reverse_path)
-        return self._path
-    path = property(_get_path)
-
-    def __hash__(self):
-        return hash(self.path)
-
-    def __eq__(self, other):
-        return self.path == other.path
-
-    def __getattr__(self, tag):
-        return getattr(self.node, tag)
-
 def option_parser():
     from calibre.utils.config import OptionParser
     parser = OptionParser(usage='%prog: [options] file')
     parser.add_option('-v', '--verbose', default=False, action='store_true',
-                      dest='verbose',
-                      help=_('Show detailed output information. Useful for debugging'))
+            dest='verbose',
+            help='Show detailed output information. Useful for debugging')
+    parser.add_option('-k', '--keep-elements', default=None, action='store',
+            dest='keep_elements',
+            help='XPath specifying elements that should not be removed')
 
     return parser
 
@@ -506,7 +490,12 @@ def main():
         raw = f.read()
 
     enc = sys.__stdout__.encoding or 'utf-8'
-    print Document(raw, default_log, debug=options.verbose).summary().encode(enc, 'replace')
+    if options.verbose:
+        default_log.filter_level = default_log.DEBUG
+    print (Document(raw, default_log,
+            debug=options.verbose,
+            keep_elements=options.keep_elements).summary().encode(enc,
+                'replace'))
 
 if __name__ == '__main__':
     main()
diff --git a/src/calibre/web/feeds/news.py b/src/calibre/web/feeds/news.py
index 436612af7e..b7efd611e0 100644
--- a/src/calibre/web/feeds/news.py
+++ b/src/calibre/web/feeds/news.py
@@ -144,6 +144,18 @@ class BasicNewsRecipe(Recipe):
     #: manually (though manual cleanup will always be superior).
     auto_cleanup = False
 
+    #: Specify elements that the auto cleanup algorithm should never remove
+    #: The syntax is a XPath expression. For example::
+    #:
+    #: auto_cleanup_keep = '//div[@id="article-image"]' will keep all divs with
+    #:                                                  id="article-image"
+    #: auto_cleanup_keep = '//*[@class="important"]' will keep all elements
+    #:                                               with class="important"
+    #: auto_cleanup_keep = '//div[@id="article-image"]|//span[@class="important"]'
+    #:                     will keep all divs with id="article-image" and spans
+    #:                     with class="important"
+    auto_cleanup_keep = None
+
     #: Specify any extra :term:`CSS` that should be addded to downloaded :term:`HTML` files
     #: It will be inserted into `<style>` tags, just before the closing
     #: `</head>` tag thereby overriding all :term:`CSS` except that which is
@@ -552,7 +564,8 @@ def extract_readable_article(self, html, url):
         from lxml.html import (fragment_fromstring, tostring,
                 document_fromstring)
 
-        doc = readability.Document(html, self.log, url=url)
+        doc = readability.Document(html, self.log, url=url,
+                keep_elements=self.auto_cleanup_keep)
         article_html = doc.summary()
         extracted_title = doc.title()