diff --git a/paddle/fluid/operators/uniform_random_op.cc b/paddle/fluid/operators/uniform_random_op.cc
index 87699362b2b5a14750a01345098ec5e6cc9be115..acaefaacdaa593c090d81084fdc1b3665314833f 100644
--- a/paddle/fluid/operators/uniform_random_op.cc
+++ b/paddle/fluid/operators/uniform_random_op.cc
@@ -24,7 +24,19 @@ template <typename T>
 class CPUUniformRandomKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    auto* tensor = ctx.Output<framework::Tensor>("Out");
+    framework::Tensor* tensor = nullptr;
+    auto out_var = ctx.OutputVar("Out");
+    if (out_var->IsType<framework::LoDTensor>()) {
+      tensor = out_var->GetMutable<framework::LoDTensor>();
+    } else if (out_var->IsType<framework::SelectedRows>()) {
+      auto shape = ctx.Attr<std::vector<int>>("shape");
+      tensor = out_var->GetMutable<framework::SelectedRows>()->mutable_value();
+      tensor->Resize(framework::make_ddim(shape));
+    } else {
+      PADDLE_THROW(
+          "uniform_random_op's output only"
+          "supports SelectedRows and Tensor");
+    }
     T* data = tensor->mutable_data<T>(ctx.GetPlace());
     unsigned int seed = static_cast<unsigned int>(ctx.Attr<int>("seed"));
     std::minstd_rand engine;
diff --git a/paddle/fluid/operators/uniform_random_op.cu b/paddle/fluid/operators/uniform_random_op.cu
index 1232cd1eb332441b12e59a34b2c2f75669925fd0..e1c7323a30233f4ec4f60e46aa6088ee6d8601b7 100644
--- a/paddle/fluid/operators/uniform_random_op.cu
+++ b/paddle/fluid/operators/uniform_random_op.cu
@@ -43,7 +43,19 @@ template <typename T>
 class GPUUniformRandomKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
-    auto* tensor = context.Output<framework::Tensor>("Out");
+    framework::Tensor* tensor = nullptr;
+    auto out_var = context.OutputVar("Out");
+    if (out_var->IsType<framework::LoDTensor>()) {
+      tensor = out_var->GetMutable<framework::LoDTensor>();
+    } else if (out_var->IsType<framework::SelectedRows>()) {
+      auto shape = context.Attr<std::vector<int>>("shape");
+      tensor = out_var->GetMutable<framework::SelectedRows>()->mutable_value();
+      tensor->Resize(framework::make_ddim(shape));
+    } else {
+      PADDLE_THROW(
+          "uniform_random_op's output only"
+          "supports SelectedRows and Tensor");
+    }
     T* data = tensor->mutable_data<T>(context.GetPlace());
     unsigned int seed = static_cast<unsigned int>(context.Attr<int>("seed"));
     if (seed == 0) {
diff --git a/python/paddle/fluid/tests/unittests/test_uniform_random_op.py b/python/paddle/fluid/tests/unittests/test_uniform_random_op.py
index 75ff85a55fc4fd504ecd032e17f7e189c17192fb..346a949b6e7c96b5535f5e65ddbada11e110a0a7 100644
--- a/python/paddle/fluid/tests/unittests/test_uniform_random_op.py
+++ b/python/paddle/fluid/tests/unittests/test_uniform_random_op.py
@@ -15,6 +15,16 @@
 import unittest
 import numpy as np
 from op_test import OpTest
+import paddle.fluid.core as core
+from paddle.fluid.op import Operator
+
+
+def output_hist(out):
+    hist, _ = np.histogram(out, range=(-5, 10))
+    hist = hist.astype("float32")
+    hist /= float(out.size)
+    prob = 0.1 * np.ones((10))
+    return hist, prob
 
 
 class TestUniformRandomOp(OpTest):
@@ -33,11 +43,37 @@ class TestUniformRandomOp(OpTest):
         self.check_output_customized(self.verify_output)
 
     def verify_output(self, outs):
-        tensor = outs[0]
-        hist, _ = np.histogram(outs[0], range=(-5, 10))
-        hist = hist.astype("float32")
-        hist /= float(outs[0].size)
-        prob = 0.1 * np.ones((10))
+        hist, prob = output_hist(np.array(outs[0]))
+        self.assertTrue(
+            np.allclose(
+                hist, prob, rtol=0, atol=0.01), "hist: " + str(hist))
+
+
+class TestUniformRandomOpSelectedRows(unittest.TestCase):
+    def get_places(self):
+        places = [core.CPUPlace()]
+        if core.is_compiled_with_cuda():
+            places.append(core.CUDAPlace(0))
+        return places
+
+    def test_check_output(self):
+        for place in self.get_places():
+            self.check_with_place(place)
+
+    def check_with_place(self, place):
+        scope = core.Scope()
+        out = scope.var("X").get_selected_rows()
+
+        op = Operator(
+            "uniform_random",
+            Out="X",
+            shape=[4, 784],
+            min=-5.0,
+            max=10.0,
+            seed=10)
+        op.run(scope, place)
+        self.assertEqual(out.get_tensor().shape(), [4, 784])
+        hist, prob = output_hist(np.array(out.get_tensor()))
         self.assertTrue(
             np.allclose(
                 hist, prob, rtol=0, atol=0.01), "hist: " + str(hist))