diff --git a/include/tim/vx/ops/elementwise.h b/include/tim/vx/ops/elementwise.h
index 905e0cb..23a758a 100644
--- a/include/tim/vx/ops/elementwise.h
+++ b/include/tim/vx/ops/elementwise.h
@@ -78,7 +78,6 @@ DECLARE_ELEMENTWISE_OP(Minimum)
 DECLARE_ELEMENTWISE_OP(Maximum)
 DECLARE_ELEMENTWISE_OP(Add)
 DECLARE_ELEMENTWISE_OP(Sub)
-DECLARE_ELEMENTWISE_OP(Div)
 DECLARE_ELEMENTWISE_OP(Pow)
 DECLARE_ELEMENTWISE_OP(FloorDiv)
 
@@ -89,6 +88,13 @@ class Multiply : public Operation {
   std::shared_ptr<Operation> Clone(std::shared_ptr<Graph>& graph) const override;
 };
 
+class Div : public Operation {
+ public:
+  Div(Graph* graph, float scale = 1.0f);
+
+  std::shared_ptr<Operation> Clone(std::shared_ptr<Graph>& graph) const override;
+};
+
 #undef DECLARE_ELEMENTWISE_OP
 
 }  // namespace ops
diff --git a/src/tim/vx/ops/elementwise.cc b/src/tim/vx/ops/elementwise.cc
index 98abcad..1589949 100644
--- a/src/tim/vx/ops/elementwise.cc
+++ b/src/tim/vx/ops/elementwise.cc
@@ -41,7 +41,6 @@ DEFINE_ELEMENTWISE_OP(Minimum, VSI_NN_OP_MINIMUM)
 DEFINE_ELEMENTWISE_OP(Maximum, VSI_NN_OP_MAXIMUM)
 DEFINE_ELEMENTWISE_OP(Add, VSI_NN_OP_ADD)
 DEFINE_ELEMENTWISE_OP(Sub, VSI_NN_OP_SUBTRACT)
-DEFINE_ELEMENTWISE_OP(Div, VSI_NN_OP_DIVIDE)
 DEFINE_ELEMENTWISE_OP(Pow, VSI_NN_OP_POW)
 DEFINE_ELEMENTWISE_OP(FloorDiv, VSI_NN_OP_FLOORDIV)
 
@@ -58,6 +57,17 @@ std::shared_ptr<Operation> Multiply::Clone(
       this->impl_->node_->nn_param.multiply.scale);
 }
 
+Div::Div(Graph* graph, float scale)
+  : Operation(graph, VSI_NN_OP_DIVIDE, 2, 1) {
+    this->impl()->node()->nn_param.divide.scale = scale;
+}
+
+std::shared_ptr<Operation> Div::Clone(
+    std::shared_ptr<Graph>& graph) const {
+  return graph->CreateOperation<Div>(
+      this->impl_->node_->nn_param.divide.scale);
+}
+
 }  // namespace ops
 }  // namespace vx
 }  // namespace tim
diff --git a/src/tim/vx/ops/elementwise_test.cc b/src/tim/vx/ops/elementwise_test.cc
index 41a904f..842d7e0 100644
--- a/src/tim/vx/ops/elementwise_test.cc
+++ b/src/tim/vx/ops/elementwise_test.cc
@@ -130,3 +130,108 @@ TEST(FloorDiv, shape_5_1_broadcast_uint8) {
     EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
     EXPECT_EQ(golden, output);
 }
+
+TEST(Div, shape_1_fp32) {
+    auto ctx = tim::vx::Context::Create();
+    auto graph = ctx->CreateGraph();
+
+    tim::vx::ShapeType io_shape({1});
+    tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
+                            io_shape, tim::vx::TensorAttribute::INPUT);
+    tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32,
+                            io_shape, tim::vx::TensorAttribute::OUTPUT);
+
+    auto input_tensor_x = graph->CreateTensor(input_spec);
+    auto input_tensor_y = graph->CreateTensor(input_spec);
+    auto output_tensor = graph->CreateTensor(output_spec);
+
+    std::vector<float> in_data_x = { 1 };
+    std::vector<float> in_data_y = { 0 };
+    std::vector<float> golden = { std::numeric_limits<float>::infinity() };
+
+    EXPECT_TRUE(input_tensor_x->CopyDataToTensor(in_data_x.data(), in_data_x.size()*4));
+    EXPECT_TRUE(input_tensor_y->CopyDataToTensor(in_data_y.data(), in_data_y.size()*4));
+    auto op = graph->CreateOperation<tim::vx::ops::Div>();
+    (*op).BindInputs({input_tensor_x, input_tensor_y}).BindOutputs({output_tensor});
+
+    EXPECT_TRUE(graph->Compile());
+    EXPECT_TRUE(graph->Run());
+    std::vector<float> output(1);
+
+    EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
+    EXPECT_EQ(golden, output);
+}
+
+TEST(Div, shape_5_1_broadcast_uint8) {
+    auto ctx = tim::vx::Context::Create();
+    auto graph = ctx->CreateGraph();
+
+    tim::vx::ShapeType in_shape_x({1});
+    tim::vx::ShapeType in_shape_y({5, 1});
+    tim::vx::ShapeType out_shape({5, 1});
+    tim::vx::Quantization quant(tim::vx::QuantType::ASYMMETRIC, 1, 0);
+    tim::vx::Quantization quant_out(tim::vx::QuantType::ASYMMETRIC, 0.5, 0);
+    tim::vx::TensorSpec input_spec_x(tim::vx::DataType::UINT8,
+                            in_shape_x, tim::vx::TensorAttribute::INPUT, quant);
+    tim::vx::TensorSpec input_spec_y(tim::vx::DataType::UINT8,
+                            in_shape_y, tim::vx::TensorAttribute::INPUT, quant);
+    tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8,
+                            out_shape, tim::vx::TensorAttribute::OUTPUT, quant_out);
+
+    auto input_tensor_x = graph->CreateTensor(input_spec_x);
+    auto input_tensor_y = graph->CreateTensor(input_spec_y);
+    auto output_tensor = graph->CreateTensor(output_spec);
+
+    std::vector<uint8_t> in_data_x = { 255 };
+    std::vector<uint8_t> in_data_y = { 1, 2, 3, 0, 255 };
+    std::vector<uint8_t> golden = { 255, 255, 170, 255, 2 };
+
+    EXPECT_TRUE(input_tensor_x->CopyDataToTensor(in_data_x.data(), in_data_x.size()));
+    EXPECT_TRUE(input_tensor_y->CopyDataToTensor(in_data_y.data(), in_data_y.size()));
+    auto op = graph->CreateOperation<tim::vx::ops::Div>();
+    (*op).BindInputs({input_tensor_x, input_tensor_y}).BindOutputs({output_tensor});
+
+    EXPECT_TRUE(graph->Compile());
+    EXPECT_TRUE(graph->Run());
+    std::vector<uint8_t> output(5);
+
+    EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
+    EXPECT_EQ(golden, output);
+}
+
+TEST(Div, shape_5_1_broadcast_scale_uint8) {
+    auto ctx = tim::vx::Context::Create();
+    auto graph = ctx->CreateGraph();
+
+    tim::vx::ShapeType in_shape_x({1});
+    tim::vx::ShapeType in_shape_y({5, 1});
+    tim::vx::ShapeType out_shape({5, 1});
+    tim::vx::Quantization quant(tim::vx::QuantType::ASYMMETRIC, 1, 0);
+    tim::vx::Quantization quant_out(tim::vx::QuantType::ASYMMETRIC, 0.5, 0);
+    tim::vx::TensorSpec input_spec_x(tim::vx::DataType::UINT8,
+                            in_shape_x, tim::vx::TensorAttribute::INPUT, quant);
+    tim::vx::TensorSpec input_spec_y(tim::vx::DataType::UINT8,
+                            in_shape_y, tim::vx::TensorAttribute::INPUT, quant);
+    tim::vx::TensorSpec output_spec(tim::vx::DataType::UINT8,
+                            out_shape, tim::vx::TensorAttribute::OUTPUT, quant_out);
+
+    auto input_tensor_x = graph->CreateTensor(input_spec_x);
+    auto input_tensor_y = graph->CreateTensor(input_spec_y);
+    auto output_tensor = graph->CreateTensor(output_spec);
+
+    std::vector<uint8_t> in_data_x = { 128 };
+    std::vector<uint8_t> in_data_y = { 1, 2, 3, 0, 255 };
+    std::vector<uint8_t> golden = { 128, 64, 43, 255, 1 };
+
+    EXPECT_TRUE(input_tensor_x->CopyDataToTensor(in_data_x.data(), in_data_x.size()));
+    EXPECT_TRUE(input_tensor_y->CopyDataToTensor(in_data_y.data(), in_data_y.size()));
+    auto op = graph->CreateOperation<tim::vx::ops::Div>(0.5f);
+    (*op).BindInputs({input_tensor_x, input_tensor_y}).BindOutputs({output_tensor});
+
+    EXPECT_TRUE(graph->Compile());
+    EXPECT_TRUE(graph->Run());
+    std::vector<uint8_t> output(5);
+
+    EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
+    EXPECT_EQ(golden, output);
+}