diff --git a/paddle/fluid/operators/reduce_ops/reduce_op.h b/paddle/fluid/operators/reduce_ops/reduce_op.h
index ecf8119ed2a191..1f1ea5c005db7f 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_op.h
+++ b/paddle/fluid/operators/reduce_ops/reduce_op.h
@@ -141,8 +141,22 @@ void HandleLargeDim(const framework::ExecutionContext& context,
 
   // transpose to 2D tensor whose shape is {unreduced, reduced}.
   const int64_t unreduced = output->numel();
-  const int64_t reduced = shuffled_input.numel() / unreduced;
+  const int64_t input_numel = shuffled_input.numel();
+  // assume: 0 / 0 == 0, which allow process 0 dim tensor
+  const int64_t reduced = (unreduced != 0) ? (input_numel / unreduced) : 0;
+
+  PADDLE_ENFORCE_EQ(
+      unreduced * reduced,
+      input_numel,
+      phi::errors::InvalidArgument(
+          "Reducing failed in HandleLargeDim, when try to transpose (%d) "
+          "operands into 2D tensor with shape (%d, %d).",
+          input_numel,
+          unreduced,
+          reduced));
+
   shuffled_input.Resize({unreduced, reduced});
+
   DDim output_dim = output->dims();
   output->Resize({unreduced});
   paddle::operators::ReduceFunctor<DeviceContext, OutT, 2, 1, Functor>(
@@ -163,7 +177,20 @@ void HandleLargeDimGrad(const framework::ExecutionContext& context,
                         Functor functor,
                         const std::vector<int>& dims) {
   const int64_t unreduced = out->numel();
-  const int64_t reduced = x->numel() / unreduced;
+  const int64_t x_numel = x->numel();
+  // assume: 0 / 0 == 0, which allow process 0 dim tensor
+  const int64_t reduced = (unreduced != 0) ? (x_numel / unreduced) : 0;
+
+  PADDLE_ENFORCE_EQ(
+      unreduced * reduced,
+      x_numel,
+      phi::errors::InvalidArgument(
+          "Reducing failed in HandleLargeDimGrad, when try to transpose (%d) "
+          "operands into 2D tensor with shape (%d, %d).",
+          x_numel,
+          unreduced,
+          reduced));
+
   DDim out_dim(out->dims());
   DDim x_dim(x->dims());
   // transpose and reshape X
diff --git a/paddle/phi/kernels/funcs/reduce_function.h b/paddle/phi/kernels/funcs/reduce_function.h
index b48f2eb4cdf2b2..0b9b852a7585d0 100644
--- a/paddle/phi/kernels/funcs/reduce_function.h
+++ b/paddle/phi/kernels/funcs/reduce_function.h
@@ -1228,8 +1228,22 @@ void HandleLargeDim(const DeviceContext& dev_ctx,
 
   // transpose to 2D tensor whose shape is {unreduced, reduced}.
   const int64_t unreduced = output->numel();
-  const int64_t reduced = shuffled_input.numel() / unreduced;
+  const int64_t input_numel = shuffled_input.numel();
+  // assume: 0 / 0 == 0, which allow process 0 dim tensor
+  const int64_t reduced = (unreduced != 0) ? (input_numel / unreduced) : 0;
+
+  PADDLE_ENFORCE_EQ(
+      unreduced * reduced,
+      input_numel,
+      phi::errors::InvalidArgument(
+          "Reducing failed in HandleLargeDim, when try to transpose (%d) "
+          "operands into 2D tensor with shape (%d, %d).",
+          input_numel,
+          unreduced,
+          reduced));
+
   shuffled_input.ResizeAndAllocate({unreduced, reduced});
+
   DDim output_dim = output->dims();
   output->ResizeAndAllocate({unreduced});
   ReduceFunctor<DeviceContext, OutT, 2, 1, Functor>(
diff --git a/paddle/phi/kernels/funcs/reduce_grad_functions.h b/paddle/phi/kernels/funcs/reduce_grad_functions.h
index 3ab7618adec48b..1b0f34b943d5af 100644
--- a/paddle/phi/kernels/funcs/reduce_grad_functions.h
+++ b/paddle/phi/kernels/funcs/reduce_grad_functions.h
@@ -87,7 +87,20 @@ void HandleLargeDimGrad(const Context& dev_ctx,
                         Functor functor,
                         const std::vector<int>& dims) {
   const int64_t unreduced = out->numel();
-  const int64_t reduced = x->numel() / unreduced;
+  const int64_t x_numel = x->numel();
+  // assume: 0 / 0 == 0, which allow process 0 dim tensor
+  const int64_t reduced = (unreduced != 0) ? (x_numel / unreduced) : 0;
+
+  PADDLE_ENFORCE_EQ(
+      unreduced * reduced,
+      x_numel,
+      phi::errors::InvalidArgument(
+          "Reducing failed in HandleLargeDimGrad, when try to transpose (%d) "
+          "operands into 2D tensor with shape (%d, %d).",
+          x_numel,
+          unreduced,
+          reduced));
+
   DDim out_dim(out->dims());
   DDim x_dim(x->dims());
   // transpose and reshape X
diff --git a/paddle/phi/kernels/reduce_min_kernel.cc b/paddle/phi/kernels/reduce_min_kernel.cc
index 660d3b753e97ea..c4c58c8342e600 100644
--- a/paddle/phi/kernels/reduce_min_kernel.cc
+++ b/paddle/phi/kernels/reduce_min_kernel.cc
@@ -26,6 +26,11 @@ void MinKernel(const Context& dev_ctx,
                bool keep_dim,
                DenseTensor* out) {
   bool reduce_all = recompute_reduce_all(x, dims);
+  PADDLE_ENFORCE_GT(
+      x.numel(),
+      0,
+      errors::InvalidArgument("Zero-size tensor to reduction operation minimum "
+                              "which has no identity."));
   MinRawKernel<T>(dev_ctx, x, dims, keep_dim, reduce_all, out);
 }
 
diff --git a/python/paddle/fluid/tests/unittests/test_min_op.py b/python/paddle/fluid/tests/unittests/test_min_op.py
index c0e9803b140f01..70d41e11321693 100644
--- a/python/paddle/fluid/tests/unittests/test_min_op.py
+++ b/python/paddle/fluid/tests/unittests/test_min_op.py
@@ -20,6 +20,7 @@
 
 import paddle
 import paddle.fluid.core as core
+from paddle import fluid
 
 
 class ApiMinTest(unittest.TestCase):
@@ -117,5 +118,18 @@ def init_data(self):
         self.keepdim = True
 
 
+class TestMinAPIWithEmptyTensor(unittest.TestCase):
+    def test_empty_tensor(self):
+        with fluid.dygraph.guard():
+            with self.assertRaises(ValueError):
+                data = np.array([], dtype=np.float32)
+                data = np.reshape(data, [0, 0, 0, 0, 0, 0, 0])
+                x = paddle.to_tensor(data, dtype='float64')
+                np_axis = np.array([0], dtype='int64')
+                tensor_axis = paddle.to_tensor(np_axis, dtype='int64')
+
+                out = paddle.min(x, tensor_axis)
+
+
 if __name__ == '__main__':
     unittest.main()