diff --git a/onnxruntime/core/providers/cpu/tensor/concat.cc b/onnxruntime/core/providers/cpu/tensor/concat.cc
index d07b3ed732..5466088553 100644
--- a/onnxruntime/core/providers/cpu/tensor/concat.cc
+++ b/onnxruntime/core/providers/cpu/tensor/concat.cc
@@ -2,6 +2,7 @@
 // Licensed under the MIT License.
 
 #include "core/providers/cpu/tensor/concat.h"
+#include "core/providers/common.h"
 
 namespace onnxruntime {
 
@@ -17,6 +18,8 @@ Status ConcatBase::PrepareForCompute(OpKernelContext* ctx, int input_count, Prep
   if (tensor_pointer == nullptr) return Status(common::ONNXRUNTIME, common::FAIL, "input count mismatch");
   const Tensor& inputs_0 = *tensor_pointer;
 
+  auto axis = HandleNegativeAxis(axis_, inputs_0.Shape().NumDimensions());
+
   // Ensure all of the non concatenated axes match each other
   for (int index = 1; index < input_count; index++) {
     tensor_pointer = ctx->Input<Tensor>(index);
@@ -25,7 +28,7 @@ Status ConcatBase::PrepareForCompute(OpKernelContext* ctx, int input_count, Prep
     // Ensure all the other axes match
     auto dimension_count = inputs_0.Shape().NumDimensions();
     for (int axis_index = 0; axis_index < dimension_count; axis_index++) {
-      if (axis_index == axis_)
+      if (axis_index == axis)
         continue;
       ONNXRUNTIME_RETURN_IF_NOT(data_n.Shape()[axis_index] == inputs_0.Shape()[axis_index], "Non concat axis dimensions must match: Axis ", axis_index, " has mismatched dimensions of ", data_n.Shape()[axis_index], " and ", inputs_0.Shape()[axis_index]);
     }
@@ -36,19 +39,19 @@ Status ConcatBase::PrepareForCompute(OpKernelContext* ctx, int input_count, Prep
   for (int index = 0; index < input_count; index++) {
     tensor_pointer = ctx->Input<Tensor>(index);
     if (tensor_pointer == nullptr) return Status(common::ONNXRUNTIME, common::FAIL, "input count mismatch");
-    concat_axis_size += tensor_pointer->Shape()[int(axis_)];
+    concat_axis_size += tensor_pointer->Shape()[int(axis)];
   }
 
   // Calculate the shape of the output tensor
   std::vector<int64_t> dims;
   for (int dimension_index = 0; dimension_index < inputs_0.Shape().NumDimensions(); dimension_index++)
     dims.emplace_back(inputs_0.Shape()[dimension_index]);
-  dims[axis_] = concat_axis_size;
+  dims[axis] = concat_axis_size;
   TensorShape outputShape(dims);
 
   // The output_axis_pitch is the number of elements to add to move to the next split axis in the output
   p.output_axis_pitch = 1;
-  for (auto i = int64_t(dims.size()); i-- > axis_;)
+  for (auto i = int64_t(dims.size()); i-- > axis;)
     p.output_axis_pitch *= dims[i];
 
   auto& concat_result = *ctx->Output(0, outputShape);
@@ -63,7 +66,7 @@ Status ConcatBase::PrepareForCompute(OpKernelContext* ctx, int input_count, Prep
 
     // The input_axis_pitch is the number of elements to add to move to the next split axis in the input
     int64_t input_axis_pitch = 1;
-    for (int i = int(data_n.Shape().NumDimensions()); i-- > axis_;)
+    for (int i = int(data_n.Shape().NumDimensions()); i-- > axis;)
       input_axis_pitch *= data_n.Shape()[i];
 
     p.inputs.push_back({&data_n, input_axis_pitch});
diff --git a/onnxruntime/test/providers/cpu/tensor/concat_op_test.cc b/onnxruntime/test/providers/cpu/tensor/concat_op_test.cc
index 8977cc9fe4..b727db53ae 100644
--- a/onnxruntime/test/providers/cpu/tensor/concat_op_test.cc
+++ b/onnxruntime/test/providers/cpu/tensor/concat_op_test.cc
@@ -29,6 +29,17 @@ TEST(MathOpTest, Concat1D_int32) {
   test.Run();
 }
 
+TEST(MathOpTest, Concat1D_int32_negative_axis) {
+  OpTester test("Concat");
+  test.AddAttribute("axis", int64_t{-1});
+
+  test.AddInput<int32_t>("input1", {1}, {1});
+  test.AddInput<int32_t>("input2", {2}, {2, 3});
+  test.AddInput<int32_t>("input3", {4}, {4, 5, 6, 7});
+  test.AddOutput<int32_t>("concat_result", {7}, {1, 2, 3, 4, 5, 6, 7});
+  test.Run();
+}
+
 TEST(MathOpTest, Concat1D) {
   OpTester test("Concat");
   test.AddAttribute("axis", int64_t{0});
@@ -103,6 +114,38 @@ TEST(MathOpTest, Concat3D_1) {
   test.Run();
 }
 
+TEST(MathOpTest, Concat3D_1_negative_axis) {
+  OpTester test("Concat");
+  test.AddAttribute("axis", int64_t{-3});
+
+  std::vector<int64_t> dims{1, 3, 3};
+  test.AddInput<float>("input1", dims,
+                       {111.0f, 112.0f, 113.0f,
+                        121.0f, 122.0f, 123.0f,
+                        131.0f, 132.0f, 133.0f});
+  test.AddInput<float>("input2", dims,
+                       {211.0f, 212.0f, 213.0f,
+                        221.0f, 222.0f, 223.0f,
+                        231.0f, 232.0f, 233.0f});
+  test.AddInput<float>("input3", dims,
+                       {311.0f, 312.0f, 313.0f,
+                        321.0f, 322.0f, 323.0f,
+                        331.0f, 332.0f, 333.0f});
+  test.AddOutput<float>("concat_result", {3, 3, 3},
+                        {111.0f, 112.0f, 113.0f,
+                         121.0f, 122.0f, 123.0f,
+                         131.0f, 132.0f, 133.0f,
+
+                         211.0f, 212.0f, 213.0f,
+                         221.0f, 222.0f, 223.0f,
+                         231.0f, 232.0f, 233.0f,
+
+                         311.0f, 312.0f, 313.0f,
+                         321.0f, 322.0f, 323.0f,
+                         331.0f, 332.0f, 333.0f});
+  test.Run();
+}
+
 TEST(MathOpTest, Concat3D_2) {
   OpTester test("Concat");
   test.AddAttribute("axis", int64_t{1});