diff --git a/jedi/evaluate/__init__.py b/jedi/evaluate/__init__.py
index 29438d31..e61b144b 100644
--- a/jedi/evaluate/__init__.py
+++ b/jedi/evaluate/__init__.py
@@ -95,8 +95,7 @@ class Evaluator(object):
 
         self.latest_grammar = parso.load_grammar(version='3.6')
         self.memoize_cache = {}  # for memoize decorators
-        # To memorize modules -> equals `sys.modules`.
-        self.modules = {}  # like `sys.modules`.
+        self.module_cache = imports.ModuleCache()  # does the job of `sys.modules`.
         self.compiled_cache = {}  # see `evaluate.compiled.create()`
         self.inferred_element_counts = {}
         self.mixed_cache = {}  # see `evaluate.compiled.mixed._create()`
diff --git a/jedi/evaluate/context/module.py b/jedi/evaluate/context/module.py
index 78d80df4..cedaed63 100644
--- a/jedi/evaluate/context/module.py
+++ b/jedi/evaluate/context/module.py
@@ -65,7 +65,7 @@ class ModuleContext(use_metaclass(CachedMetaClass, TreeContext)):
 
     # I'm not sure if the star import cache is really that effective anymore
     # with all the other really fast import caches. Recheck. Also we would need
-    # to push the star imports into Evaluator.modules, if we reenable this.
+    # to push the star imports into Evaluator.module_cache, if we reenable this.
     @evaluator_method_cache([])
     def star_imports(self):
         modules = []
@@ -115,7 +115,7 @@ class ModuleContext(use_metaclass(CachedMetaClass, TreeContext)):
         return None
 
     def py__name__(self):
-        for name, module in self.evaluator.modules.items():
+        for name, module in self.evaluator.module_cache.iterate_modules_with_names():
             if module == self and name != '':
                 return name
 
diff --git a/jedi/evaluate/imports.py b/jedi/evaluate/imports.py
index 586458a6..5f976b49 100644
--- a/jedi/evaluate/imports.py
+++ b/jedi/evaluate/imports.py
@@ -31,6 +31,23 @@ from jedi.evaluate.filters import AbstractNameDefinition
 from jedi.evaluate.base_context import ContextSet, NO_CONTEXTS
 
 
+class ModuleCache():
+    def __init__(self):
+        self._path_cache = {}
+        self._name_cache = {}
+
+    def add(self, module, name):
+        path = module.py__file__()
+        self._path_cache[path] = module
+        self._name_cache[name] = module
+
+    def iterate_modules_with_names(self):
+        return self._name_cache.items()
+
+    def get(self, name):
+        return self._name_cache[name]
+
+
 # This memoization is needed, because otherwise we will infinitely loop on
 # certain imports.
 @evaluator_method_cache(default=NO_CONTEXTS)
@@ -289,7 +306,7 @@ class Importer(object):
 
         module_name = '.'.join(import_parts)
         try:
-            return ContextSet(self._evaluator.modules[module_name])
+            return ContextSet(self._evaluator.module_cache.get(module_name))
         except KeyError:
             pass
 
@@ -367,7 +384,7 @@ class Importer(object):
             # importable.
             return NO_CONTEXTS
 
-        self._evaluator.modules[module_name] = module
+        self._evaluator.module_cache.add(module, module_name)
         return ContextSet(module)
 
     def _generate_name(self, name, in_module=None):
@@ -481,7 +498,7 @@ def add_module(evaluator, module_name, module):
         # the sepatator dots for nested packages. Therefore we return
         # `__main__` in ModuleWrapper.py__name__(), which is similar to
         # Python behavior.
-        evaluator.modules[module_name] = module
+        evaluator.module_cache.add(module, module_name)
 
 
 def get_modules_containing_name(evaluator, modules, name):