diff -r ce4bb37a1409 sphinx/domains/std.py
--- a/sphinx/domains/std.py	Sat Feb 05 21:48:54 2011 +0100
+++ b/sphinx/domains/std.py	Wed Mar 16 14:10:54 2011 -0300
@@ -511,8 +511,13 @@
                                                                      ('','',''))
             if not docname:
                 return None
-            newnode = nodes.reference('', '', internal=True)
-            innernode = nodes.emphasis(sectname, sectname)
+            newnode = nodes.reference('', '', internal=True)            
+     
+            if labelid.endswith('-num') and builder.name == 'latex':
+                innernode = nodes.raw('', '') 
+            else:
+                innernode = nodes.emphasis(sectname, sectname) 
+
             if docname == fromdocname:
                 newnode['refid'] = labelid
             else:
diff -r ce4bb37a1409 sphinx/writers/latex.py
--- a/sphinx/writers/latex.py	Sat Feb 05 21:48:54 2011 +0100
+++ b/sphinx/writers/latex.py	Wed Mar 16 14:10:54 2011 -0300
@@ -275,6 +275,9 @@
     def idescape(self, id):
         return str(unicode(id).translate(tex_replace_map))
 
+    def reffig(self, id):
+        return '~\\ref{%s}' % self.idescape(id)
+
     def generate_indices(self):
         def generate(content, collapsed):
             ret.append('\\begin{theindex}\n')
@@ -577,6 +580,7 @@
 
     def visit_refcount(self, node):
         self.body.append("\\emph{")
+
     def depart_refcount(self, node):
         self.body.append("}\\\\")
 
@@ -1141,10 +1145,17 @@
             if hashindex == -1:
                 # reference to the document
                 id = uri[1:] + '::doc'
+                self.body.append(self.hyperlink(id))
             else:
                 # reference to a label
                 id = uri[1:].replace('#', ':')
-            self.body.append(self.hyperlink(id))
+                if id.endswith('-num'):
+                    self.body.append(self.reffig(id))
+                    self.context.append('')
+                    return
+                else:
+                    self.body.append(self.hyperlink(id))
+                    
             if len(node) and hasattr(node[0], 'attributes') and \
                    'std-term' in node[0].get('classes', []):
                 # don't add a pageref for glossary terms