diff --git a/pymnn/pip_package/MNN/numpy/__init__.py b/pymnn/pip_package/MNN/numpy/__init__.py
index a9ea5c390c..dfd78c4857 100644
--- a/pymnn/pip_package/MNN/numpy/__init__.py
+++ b/pymnn/pip_package/MNN/numpy/__init__.py
@@ -1301,8 +1301,8 @@ def quantile(a, q, axis=None, out=None, overwrite_input=False, interpolation='li
 def median(a, axis=None, out=None, overwrite_input=False, keepdims=False):__not_impl()
 def mean(a, axis=None, dtype=float32, out=None, keepdims=False):
     a = _F.cast(a, dtype)
-    if axis is None: return asscalar(_F.reduce_mean(ravel(a)))
-    return _F.reduce_mean(a, axis, keepdims)
+    if axis is None and not keepdims: return asscalar(_F.reduce_mean(ravel(a)))
+    return _F.reduce_mean(a, [] if axis is None else axis, keepdims)
 def average(a, axis=None, weights=None, returned=False):
     if weights is not None:
         raise ValueError('MNN.numpy average not support `weights`.')
diff --git a/pymnn/test/unit_test.py b/pymnn/test/unit_test.py
index 3f118362ff..71132c74b1 100644
--- a/pymnn/test/unit_test.py
+++ b/pymnn/test/unit_test.py
@@ -1187,6 +1187,19 @@ def test_searching_counting(self):
         self.assertEqualVar(mp.flatnonzero(x), np.flatnonzero(x_))
         self.assertEqualVar(mp.count_nonzero(x, 0), np.count_nonzero(x_, 0))
     def test_statistics(self):
+        for shape in [(3,), (2, 3), (2, 3, 2), (2, 1, 3, 2)]:
+            data = (np.arange(np.prod(shape), dtype=np.float32).reshape(shape) % 7 - 3) * 0.5
+            for axis in [None] + list(range(len(shape))):
+                for keepdims in [False, True]:
+                    with self.subTest(shape=shape, axis=axis, keepdims=keepdims):
+                        value = mp.array(data)
+                        actual = mp.mean(value, axis=axis, keepdims=keepdims)
+                        expected = np.mean(data, axis=axis, keepdims=keepdims)
+                        if keepdims or axis is not None:
+                            self.assertEqualVar(actual, expected)
+                        else:
+                            self.assertAlmostEqual(actual, float(expected), places=5)
+                        self.assertEqualVar(value, data)
         x = mp.array([[1,2,3], [4,6,5]])
         x_ = np.array([[1,2,3], [4,6,5]])
         self.assertEqual(mp.amin(x), np.amin(x_))