diff --git a/scrapely/extraction/__init__.py b/scrapely/extraction/__init__.py
index a4c10c4..99c6d67 100644
--- a/scrapely/extraction/__init__.py
+++ b/scrapely/extraction/__init__.py
@@ -35,6 +35,7 @@ class InstanceBasedLearningExtractor(object):
         RepeatedDataExtractor,
         RecordExtractor,
     ]
+    _ext_items_max_number = 2
 
     def __init__(self, td_pairs, trace=False, apply_extrarequired=True):
         """Initialise this extractor
@@ -77,11 +78,11 @@ def __init__(self, td_pairs, trace=False, apply_extrarequired=True):
             modified_parsed_tdpairs.append((parsed, (t, descriptor)))
         # templates with more attributes are considered first
         sorted_tdpairs = sorted(modified_parsed_tdpairs,
-                key=lambda x: _annotation_count(x[0]), reverse=True)
+                                key=lambda x: _annotation_count(x[0]), reverse=True)
         self.extraction_trees = [
             self.build_extraction_tree(p, td[1], trace)
             for p, td in sorted_tdpairs
-        ]
+            ]
         self.validated = dict(
             (td[0].page_id, td[1].validated if td[1] else self._filter_not_none)
             for _, td in sorted_tdpairs
@@ -108,23 +109,27 @@ def extract(self, html, pref_template_id=None):
         If pref_template_url is specified, the template with that url will be
         used first.
         """
+        max_extracted_value = {}
+        correctly_extracted_template = ''
         extraction_page = parse_extraction_page(self.token_dict, html)
         if pref_template_id is not None:
             extraction_trees = sorted(self.extraction_trees,
-                    key=lambda x: x.template.id != pref_template_id)
+                                      key=lambda x: x.template.id != pref_template_id)
         else:
             extraction_trees = self.extraction_trees
 
         for extraction_tree in extraction_trees:
             extracted = extraction_tree.extract(extraction_page)
             correctly_extracted = self.validated[extraction_tree.template.id](extracted)
-            if len(correctly_extracted) > 0:
-                return correctly_extracted, extraction_tree.template
-        return None, None
+            if len(correctly_extracted[0]) > len(max_extracted_value):
+                max_extracted_value = correctly_extracted[0]
+                correctly_extracted_template = extraction_tree.template
+
+        return [max_extracted_value], correctly_extracted_template
 
     def __str__(self):
         return "InstanceBasedLearningExtractor[\n%s\n]" % \
-                (',\n'.join(map(str, self.extraction_trees)))
+               (',\n'.join(map(str, self.extraction_trees)))
 
     @staticmethod
     def _filter_not_none(items):