diff --git a/IPython/testing/plugin/ipdoctest.py b/IPython/testing/plugin/ipdoctest.py
index 3743a08..4380f4b 100644
--- a/IPython/testing/plugin/ipdoctest.py
+++ b/IPython/testing/plugin/ipdoctest.py
@@ -439,7 +439,29 @@ class DocTestCase(doctests.DocTestCase):
             _ip.IP.user_ns.update(self._dt_test.globs)
             self._dt_test.globs = _ip.IP.user_ns
 
-        doctests.DocTestCase.setUp(self)
+        super(DocTestCase, self).setUp()
+
+    def tearDown(self):
+        # XXX - fperez: I am not sure if this is truly a bug in nose 0.11, but
+        # it does look like one to me: its tearDown method tries to run
+        #
+        # delattr(__builtin__, self._result_var)
+        #
+        # without checking that the attribute really is there; it implicitly
+        # assumes it should have been set via displayhook.  But if the
+        # displayhook was never called, this doesn't necessarily happen.  I
+        # haven't been able to find a little self-contained example outside of
+        # ipython that would show the problem so I can report it to the nose
+        # team, but it does happen a lot in our code.
+        #
+        # So here, we just protect as narrowly as possible by trapping an
+        # attribute error whose message would be the name of self._result_var,
+        # and letting any other error propagate.
+        try:
+            super(DocTestCase, self).tearDown()
+        except AttributeError, exc:
+            if exc.message != self._result_var:
+                raise
 
 
 # A simple subclassing of the original with a different class name, so we can