diff --git a/imperative/python/megengine/autodiff/grad_manager.py b/imperative/python/megengine/autodiff/grad_manager.py
index db03f235..0cd44b38 100644
--- a/imperative/python/megengine/autodiff/grad_manager.py
+++ b/imperative/python/megengine/autodiff/grad_manager.py
@@ -115,14 +115,6 @@ class GradManager:
             else:
                 logger.warning("params with index {} is not attached.".format(idx))
 
-    def clear_grad(self):
-        r"""
-        For advanced usage: set the grad attribute to None for registered parameters.
-        It could be more convenient when there is more than one Optimizer.
-        """
-        for param in self._param_dict.values():
-            param.grad = None
-
     def _register_after_backward_callback(self, callback):
         self._after_backward_callback.append(callback)
         return self
diff --git a/imperative/python/test/unit/autodiff/test_grad_manger.py b/imperative/python/test/unit/autodiff/test_grad_manger.py
index f47e618b..947fa520 100644
--- a/imperative/python/test/unit/autodiff/test_grad_manger.py
+++ b/imperative/python/test/unit/autodiff/test_grad_manger.py
@@ -37,7 +37,8 @@ def test_basic():
     np.testing.assert_equal(w.grad.numpy(), [[1], [3], [5]])
     np.testing.assert_equal(b.grad.numpy(), [1])
 
-    gm.clear_grad()
+    w.grad = None
+    b.grad = None
     with gm:
         p = F.matmul(x, w)
         y = p + b