diff --git a/Include/pythonrun.h b/Include/pythonrun.h
--- a/Include/pythonrun.h
+++ b/Include/pythonrun.h
@@ -217,6 +217,7 @@ PyAPI_FUNC(void) PyBytes_Fini(void);
 PyAPI_FUNC(void) PyByteArray_Fini(void);
 PyAPI_FUNC(void) PyFloat_Fini(void);
 PyAPI_FUNC(void) PyOS_FiniInterrupts(void);
+PyAPI_FUNC(void) _PyGC_DumpShutdownStats(void);
 PyAPI_FUNC(void) _PyGC_Fini(void);
 PyAPI_FUNC(void) PySlice_Fini(void);
 PyAPI_FUNC(void) _PyType_Fini(void);
diff --git a/Lib/test/test_gc.py b/Lib/test/test_gc.py
--- a/Lib/test/test_gc.py
+++ b/Lib/test/test_gc.py
@@ -1,6 +1,8 @@
 import unittest
 from test.support import (verbose, refcount_test, run_unittest,
                             strip_python_stderr)
+from test.script_helper import assert_python_ok, make_script, temp_dir
+
 import sys
 import time
 import gc
@@ -610,6 +612,40 @@ class GCTests(unittest.TestCase):
         stderr = run_command(code % "gc.DEBUG_SAVEALL")
         self.assertNotIn(b"uncollectable objects at shutdown", stderr)
 
+    def test_gc_main_module_at_shutdown(self):
+        # Create a reference cycle through the __main__ module and check
+        # it gets collected at interpreter shutdown.
+        code = """if 1:
+            import weakref
+            class C:
+                def __del__(self):
+                    print('__del__ called')
+            l = [C()]
+            l.append(l)
+            """
+        rc, out, err = assert_python_ok('-c', code)
+        self.assertEqual(out.strip(), b'__del__ called')
+
+    def test_gc_ordinary_module_at_shutdown(self):
+        # Same as above, but with a non-__main__ module.
+        with temp_dir() as script_dir:
+            module = """if 1:
+                import weakref
+                class C:
+                    def __del__(self):
+                        print('__del__ called')
+                l = [C()]
+                l.append(l)
+                """
+            code = """if 1:
+                import sys
+                sys.path.insert(0, %r)
+                import gctest
+                """ % (script_dir,)
+            make_script(script_dir, 'gctest', module)
+            rc, out, err = assert_python_ok('-c', code)
+            self.assertEqual(out.strip(), b'__del__ called')
+
     def test_get_stats(self):
         stats = gc.get_stats()
         self.assertEqual(len(stats), 3)
diff --git a/Modules/gcmodule.c b/Modules/gcmodule.c
--- a/Modules/gcmodule.c
+++ b/Modules/gcmodule.c
@@ -1544,8 +1544,9 @@ PyGC_Collect(void)
     return n;
 }
 
+
 void
-_PyGC_Fini(void)
+_PyGC_DumpShutdownStats(void)
 {
     if (!(debug & DEBUG_SAVEALL)
         && garbage != NULL && PyList_GET_SIZE(garbage) > 0) {
@@ -1574,6 +1575,11 @@ void
             Py_XDECREF(bytes);
         }
     }
+}
+
+void
+_PyGC_Fini(void)
+{
     Py_CLEAR(callbacks);
 }
 
diff --git a/Python/import.c b/Python/import.c
--- a/Python/import.c
+++ b/Python/import.c
@@ -403,6 +403,14 @@ PyImport_Cleanup(void)
         }
     }
 
+    /* Collect garbage remaining after deleting the modules. Mostly
+       reference cycles created by classes. */
+    PyGC_Collect();
+
+    /* Dump GC stats before it's too late, since it uses the warnings
+       machinery. */
+    _PyGC_DumpShutdownStats();
+
     /* Next, delete sys and builtins (in that order) */
     value = PyDict_GetItemString(modules, "sys");
     if (value != NULL && PyModule_Check(value)) {
diff --git a/Python/pythonrun.c b/Python/pythonrun.c
--- a/Python/pythonrun.c
+++ b/Python/pythonrun.c
@@ -544,10 +544,6 @@ Py_Finalize(void)
     while (PyGC_Collect() > 0)
         /* nothing */;
 #endif
-    /* We run this while most interpreter state is still alive, so that
-       debug information can be printed out */
-    _PyGC_Fini();
-
     /* Destroy all modules */
     PyImport_Cleanup();
 
@@ -596,7 +592,7 @@ Py_Finalize(void)
         _Py_PrintReferences(stderr);
 #endif /* Py_TRACE_REFS */
 
-    /* Clear interpreter state */
+    /* Clear interpreter state and all thread states */
     PyInterpreterState_Clear(interp);
 
     /* Now we decref the exception classes.  After this point nothing
@@ -612,10 +608,6 @@ Py_Finalize(void)
     _PyGILState_Fini();
 #endif /* WITH_THREAD */
 
-    /* Delete current thread */
-    PyThreadState_Swap(NULL);
-    PyInterpreterState_Delete(interp);
-
     /* Sundry finalizers */
     PyMethod_Fini();
     PyFrame_Fini();
@@ -630,10 +622,15 @@ Py_Finalize(void)
     PyDict_Fini();
     PySlice_Fini();
     _PyType_Fini();
+    _PyGC_Fini();
 
     /* Cleanup Unicode implementation */
     _PyUnicode_Fini();
 
+    /* Delete current thread, only after all Python code has executed */
+    PyThreadState_Swap(NULL);
+    PyInterpreterState_Delete(interp);
+
     /* reset file system default encoding */
     if (!Py_HasFileSystemDefaultEncoding && Py_FileSystemDefaultEncoding) {
         free((char*)Py_FileSystemDefaultEncoding);