diff --git a/mindspore/ops/_grad/grad_nn_ops.py b/mindspore/ops/_grad/grad_nn_ops.py
index de93d718502..9e6a8791afb 100755
--- a/mindspore/ops/_grad/grad_nn_ops.py
+++ b/mindspore/ops/_grad/grad_nn_ops.py
@@ -930,7 +930,7 @@ def get_bprop_kl_div_loss(self):
 @bprop_getters.register(P.Dropout)
 def get_bprop_dropout(self):
     """Grad definition for `Dropout` operation."""
-    grad = P.DropoutGrad(self.keep_prob)
+    grad = G.DropoutGrad(self.keep_prob)
 
     def bprop(x, out, dout):
         _, mask = out
diff --git a/mindspore/ops/operations/__init__.py b/mindspore/ops/operations/__init__.py
index 396ece17239..9bc167d91a5 100644
--- a/mindspore/ops/operations/__init__.py
+++ b/mindspore/ops/operations/__init__.py
@@ -61,7 +61,7 @@ from .random_ops import (RandomChoiceWithMask, StandardNormal, Gamma, Poisson, U
 from .nn_ops import (LSTM, SGD, Adam, FusedSparseAdam, FusedSparseLazyAdam, ApplyMomentum, BatchNorm,
                      BiasAdd, Conv2D,
                      DepthwiseConv2dNative,
-                     DropoutDoMask, DropoutGrad, Dropout,
+                     DropoutDoMask, Dropout,
                      DropoutGenMask, Flatten, FusedBatchNorm, FusedBatchNormEx, BNTrainingReduce, BNTrainingUpdate,
                      Gelu, Elu,
                      GetNext, L2Normalize, LayerNorm, L2Loss, CTCLoss, CTCLossV2, CTCGreedyDecoder,
@@ -211,7 +211,6 @@ __all__ = [
     'DynamicShape',
     'DropoutDoMask',
     'DropoutGenMask',
-    'DropoutGrad',
     'Dropout',
     'Neg',
     'InplaceAdd',
diff --git a/mindspore/ops/operations/_grad_ops.py b/mindspore/ops/operations/_grad_ops.py
index af103f0bd5f..b408419c297 100644
--- a/mindspore/ops/operations/_grad_ops.py
+++ b/mindspore/ops/operations/_grad_ops.py
@@ -462,6 +462,42 @@ class DepthwiseConv2dNativeBackpropInput(PrimitiveWithInfer):
         return out
 
 
+class DropoutGrad(PrimitiveWithInfer):
+    """
+    The gradient of Dropout. During training, randomly zeroes some of the elements
+    of the input tensor with probability.
+
+    Args:
+        keep_prob (float): The keep rate, between 0 and 1, e.g. keep_prob = 0.9,
+          means dropping out 10% of input units.
+
+    Inputs:
+        - **shape** (tuple[int]) - The shape of target mask.
+
+    Outputs:
+        Tensor, the value of generated mask for input shape.
+
+    Examples:
+        >>> dropout_grad = P.DropoutGrad(keep_prob=0.5)
+        >>> in = Tensor((20, 16, 50, 50))
+        >>> out = dropout_grad(in)
+    """
+
+    @prim_attr_register
+    def __init__(self, keep_prob=0.5):
+        self.keep_prob = validator.check_number_range("keep_prob", keep_prob, 0, 1, Rel.INC_RIGHT, self.name)
+
+    def infer_shape(self, dy_shape, mask_shape):
+        return dy_shape
+
+    def infer_dtype(self, dy_dtype, mask_dtype):
+        valid_types = (mstype.float16, mstype.float32)
+        validator.check_subclass("dy", dy_dtype, mstype.tensor, self.name)
+        validator.check_subclass("mask", mask_dtype, mstype.tensor, self.name)
+        validator.check_tensor_type_same({"dy_dtype": dy_dtype}, valid_types, self.name)
+        return dy_dtype
+
+
 class FlattenGrad(PrimitiveWithInfer):
     """Performs gradients of Flatten."""
 
diff --git a/mindspore/ops/operations/nn_ops.py b/mindspore/ops/operations/nn_ops.py
index 11b674dc079..641b3a4f6cb 100644
--- a/mindspore/ops/operations/nn_ops.py
+++ b/mindspore/ops/operations/nn_ops.py
@@ -5247,42 +5247,6 @@ class Dropout(PrimitiveWithInfer):
         return x_dtype, x_dtype
 
 
-class DropoutGrad(PrimitiveWithInfer):
-    """
-    The gradient of Dropout. During training, randomly zeroes some of the elements
-    of the input tensor with probability.
-
-    Args:
-        keep_prob (float): The keep rate, between 0 and 1, e.g. keep_prob = 0.9,
-          means dropping out 10% of input units.
-
-    Inputs:
-        - **shape** (tuple[int]) - The shape of target mask.
-
-    Outputs:
-        Tensor, the value of generated mask for input shape.
-
-    Examples:
-        >>> dropout_grad = P.DropoutGrad(keep_prob=0.5)
-        >>> in = Tensor((20, 16, 50, 50))
-        >>> out = dropout_grad(in)
-    """
-
-    @prim_attr_register
-    def __init__(self, keep_prob=0.5):
-        self.keep_prob = validator.check_number_range("keep_prob", keep_prob, 0, 1, Rel.INC_RIGHT, self.name)
-
-    def infer_shape(self, dy_shape, mask_shape):
-        return dy_shape
-
-    def infer_dtype(self, dy_dtype, mask_dtype):
-        valid_types = (mstype.float16, mstype.float32)
-        validator.check_subclass("dy", dy_dtype, mstype.tensor, self.name)
-        validator.check_subclass("mask", mask_dtype, mstype.tensor, self.name)
-        validator.check_tensor_type_same({"dy_dtype": dy_dtype}, valid_types, self.name)
-        return dy_dtype
-
-
 class CTCLoss(PrimitiveWithInfer):
     """
     Calculates the CTC (Connectionist Temporal Classification) loss and the gradient.