diff --git a/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_builder.cc b/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_builder.cc
index 319ebe26c5..a351a5c483 100644
--- a/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_builder.cc
+++ b/onnxruntime/core/providers/nnapi/nnapi_builtin/builders/op_builder.cc
@@ -1284,8 +1284,10 @@ Status PoolOpBuilder::AddToModelBuilderImpl(ModelBuilder& model_builder, const N
   int32_t fuse_code = model_builder.FindActivation(node, *node.OutputDefs()[0]);
 
   // Get output scale and zero point if this is QLinearAveragePool
-  float y_scale = 0.0f;
-  int32_t y_zero_point = 0;
+  // Otherwise we will use the scale and zero point of the input
+  const OperandType& input_operand_type = operand_types.at(input);
+  float y_scale = input_operand_type.operandType.scale;
+  int32_t y_zero_point = input_operand_type.operandType.zeroPoint;
   if (is_qlinear_average_pool) {
     const auto& initializers = model_builder.GetInitializerTensors();
     float x_scale = 0.0f;
diff --git a/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc b/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
index b80aa7cb85..b8c4acd127 100644
--- a/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
+++ b/onnxruntime/test/providers/nnapi/nnapi_basic_test.cc
@@ -82,7 +82,7 @@ TEST(NnapiExecutionProviderTest, InternalUint8SupportTest) {
   const ORTCHAR_T* model_file_name = ORT_TSTR("testdata/nnapi_internal_uint8_support.onnx");
 
 #if defined(__ANDROID__)
-  std::vector<int64_t> dims_x = {1, 3};
+  std::vector<int64_t> dims_x = {1, 1, 1, 3};
   std::vector<float> values_x = {0.0f, 256.0f, 512.0f};
   OrtValue ml_value_x;
   CreateMLValue<float>(TestNnapiExecutionProvider()->GetAllocator(0, OrtMemTypeDefault), dims_x, values_x,
diff --git a/onnxruntime/test/testdata/nnapi_internal_uint8_support.onnx b/onnxruntime/test/testdata/nnapi_internal_uint8_support.onnx
index daee608f7c..03fca2f533 100644
Binary files a/onnxruntime/test/testdata/nnapi_internal_uint8_support.onnx and b/onnxruntime/test/testdata/nnapi_internal_uint8_support.onnx differ
diff --git a/onnxruntime/test/testdata/nnapi_internal_uint8_support.py b/onnxruntime/test/testdata/nnapi_internal_uint8_support.py
index f419df2dc7..d7d2c18cf5 100644
--- a/onnxruntime/test/testdata/nnapi_internal_uint8_support.py
+++ b/onnxruntime/test/testdata/nnapi_internal_uint8_support.py
@@ -8,10 +8,11 @@ from onnx import TensorProto
 # def GenerateModel(model_name):
 def GenerateModel(model_name):
     nodes = [
-        helper.make_node("QuantizeLinear", ["X", "Scale", "Zero_point"], ["X_quantized"], "quantize"),
-        helper.make_node("Concat", ["X_quantized", "X_quantized"], ["X_concat"], axis=0, name="concat"),
-        helper.make_node("Transpose", ["X_concat"], ["X_transposed"], "transpose"),
-        helper.make_node("DequantizeLinear", ["X_transposed", "Scale", "Zero_point"], ["Y"], "dequantize"),
+        helper.make_node("QuantizeLinear", ["X", "Scale", "Zero_point"], ["X_quantized"], "quantize_0"),
+        helper.make_node("Concat", ["X_quantized", "X_quantized"], ["X_concat"], axis=-2, name="concat_0"),
+        helper.make_node("MaxPool", ["X_concat"], ["X_maxpool"], kernel_shape=[2, 2], name="maxpool_0"),
+        helper.make_node("Transpose", ["X_maxpool"], ["X_transposed"], perm=[0, 1, 3, 2], name="transpose_0"),
+        helper.make_node("DequantizeLinear", ["X_transposed", "Scale", "Zero_point"], ["Y"], "dequantize_0"),
     ]
 
     initializers = [
@@ -20,14 +21,14 @@ def GenerateModel(model_name):
     ]
 
     inputs = [
-        helper.make_tensor_value_info('X', TensorProto.FLOAT, [1, 3]),
+        helper.make_tensor_value_info('X', TensorProto.FLOAT, [1, 1, 1, 3]),
     ]
 
     graph = helper.make_graph(
         nodes,
         "NNAPI_Internal_uint8_Test",
         inputs,
-        [helper.make_tensor_value_info('Y', TensorProto.FLOAT, [3, 2])],
+        [helper.make_tensor_value_info('Y', TensorProto.FLOAT, [1, 1, 2, 1])],
         initializers
     )