diff --git a/jedi/evaluate/analysis.py b/jedi/evaluate/analysis.py
index 7c99796c..ef823381 100644
--- a/jedi/evaluate/analysis.py
+++ b/jedi/evaluate/analysis.py
@@ -19,6 +19,7 @@ CODES = {
     'type-error-star': (10, TypeError, None),
     'type-error-operation': (11, TypeError, None),
     'type-error-not-iterable': (12, TypeError, None),
+    'type-error-isinstance': (13, TypeError, None),
 }
 
 
diff --git a/jedi/evaluate/compiled/__init__.py b/jedi/evaluate/compiled/__init__.py
index 88555543..b5b5ecba 100644
--- a/jedi/evaluate/compiled/__init__.py
+++ b/jedi/evaluate/compiled/__init__.py
@@ -208,8 +208,8 @@ class CompiledObject(Base):
                 # Get rid of side effects, we won't call custom `__getitem__`s.
                 return
 
-            for obj in self.obj:
-                yield set([CompiledObject(obj)])
+            for part in self.obj:
+                yield set([CompiledObject(part)])
         return actual
 
     @property
diff --git a/jedi/evaluate/param.py b/jedi/evaluate/param.py
index e33f2f3e..feef396d 100644
--- a/jedi/evaluate/param.py
+++ b/jedi/evaluate/param.py
@@ -12,8 +12,13 @@ from jedi.evaluate.helpers import FakeName
 from jedi.cache import underscore_memoization
 
 
-def try_iter_content(types):
+def try_iter_content(types, depth=0):
     """Helper method for static analysis."""
+    if depth > 10:
+        # It's possible that a loop has references on itself (especially with
+        # CompiledObject). Therefore don't loop infinitely.
+        return
+
     for typ in types:
         try:
             f = typ.py__iter__
@@ -21,7 +26,7 @@ def try_iter_content(types):
             pass
         else:
             for iter_types in f():
-                try_iter_content(iter_types)
+                try_iter_content(iter_types, depth + 1)
 
 
 class Arguments(tree.Base):
diff --git a/jedi/evaluate/stdlib.py b/jedi/evaluate/stdlib.py
index c658448c..98d5351b 100644
--- a/jedi/evaluate/stdlib.py
+++ b/jedi/evaluate/stdlib.py
@@ -19,6 +19,7 @@ from jedi.parser import tree
 from jedi import debug
 from jedi.evaluate import precedence
 from jedi.evaluate import param
+from jedi.evaluate import analysis
 
 
 class NotInStdLib(LookupError):
@@ -40,9 +41,11 @@ def execute(evaluator, obj, arguments):
 
         # for now we just support builtin functions.
         try:
-            return _implemented[module_name][obj_name](evaluator, obj, arguments)
+            func = _implemented[module_name][obj_name]
         except KeyError:
             pass
+        else:
+            return func(evaluator, obj, arguments)
     raise NotInStdLib()
 
 
@@ -186,12 +189,15 @@ def builtins_isinstance(evaluator, objects, types, arguments):
         for cls_or_tup in types:
             if cls_or_tup.is_class():
                 bool_results.add(cls_or_tup in mro)
-            else:
-                # TODO Actually we should check for non iterables here (the
-                # type should be object or tuple of object)
+            elif str(cls_or_tup.name) == 'tuple' \
+                    and cls_or_tup.get_parent_scope() == compiled.builtin:
                 # Check for tuples.
-                classes = iterable.py__iter__types(evaluator, set([cls_or_tup]))
+                classes = unite(cls_or_tup.py__iter__())
                 bool_results.add(any(cls in mro for cls in classes))
+            else:
+                _, nodes = list(arguments.unpack())[1]
+                for node in nodes:
+                    analysis.add(evaluator, 'type-error-isinstance', node)
 
     return set(compiled.keyword_from_value(x) for x in bool_results)