diff --git a/onnxruntime/core/providers/cpu/tensor/upsample.h b/onnxruntime/core/providers/cpu/tensor/upsample.h
index 8ff04781f6ad0..c23347a17febc 100644
--- a/onnxruntime/core/providers/cpu/tensor/upsample.h
+++ b/onnxruntime/core/providers/cpu/tensor/upsample.h
@@ -125,10 +125,17 @@ void UpsampleBilinear(const int32_t batch_size,
               T X12 = Xdata[p.input_width_mul_y2[y] + p.in_x1[x]];
               T X22 = Xdata[p.input_width_mul_y2[y] + p.in_x2[x]];
 
-              Ydata[output_offset] = static_cast<T>(p.dx2[x] * p.dy2[y] * X11 +
-                                                    p.dx1[x] * p.dy2[y] * X21 +
-                                                    p.dx2[x] * p.dy1[y] * X12 +
-                                                    p.dx1[x] * p.dy1[y] * X22);
+              if constexpr (is_floating_point_v<T>) {
+                Ydata[output_offset] = static_cast<T>(p.dx2[x] * p.dy2[y] * X11 +
+                                                      p.dx1[x] * p.dy2[y] * X21 +
+                                                      p.dx2[x] * p.dy1[y] * X12 +
+                                                      p.dx1[x] * p.dy1[y] * X22);
+              } else {
+                Ydata[output_offset] = static_cast<T>(nearbyintf(p.dx2[x] * p.dy2[y] * X11 +
+                                                                 p.dx1[x] * p.dy2[y] * X21 +
+                                                                 p.dx2[x] * p.dy1[y] * X12 +
+                                                                 p.dx1[x] * p.dy1[y] * X22));
+              }
             }
           }
         });
diff --git a/onnxruntime/test/optimizer/qdq_transformer_test.cc b/onnxruntime/test/optimizer/qdq_transformer_test.cc
index d07977d4b97b8..fbcb1cf942f86 100644
--- a/onnxruntime/test/optimizer/qdq_transformer_test.cc
+++ b/onnxruntime/test/optimizer/qdq_transformer_test.cc
@@ -1924,6 +1924,52 @@ TEST(QDQTransformerTests, Resize) {
   test_case({2, 13, 12, 37}, rand_gen.Uniform<int64_t>(std::vector<int64_t>{4}, 1, 16), true /*use_contrib_qdq*/);
 }
 
+// Regression test for a model with a Resize node using QDQ quantization.
+// Tests handling of rounding issue inconsistency between int Resize and Resize with QDQ quantization.
+// See https://github.com/microsoft/onnxruntime/issues/21319 for more details.
+TEST(QDQTransformerTests, RegressionTest_GitHubIssue21319) {
+  Status status;
+  auto model_uri = ORT_TSTR("testdata/qdq_transformer/gh_issue_21319.onnx");
+
+  std::vector<int64_t> input0_dims{4};
+  std::vector<int64_t> input1_dims{1, 1, 1, 2};
+  std::vector<float> input0_data = {1.0, 1.0, 1.0, 2.0};
+  std::vector<float> input1_data = {0.0, 1.0};
+
+  OrtValue input0;
+  OrtValue input1;
+  CreateMLValue<float>(TestCPUExecutionProvider()->CreatePreferredAllocators()[0], input0_dims, input0_data, &input0);
+  CreateMLValue<float>(TestCPUExecutionProvider()->CreatePreferredAllocators()[0], input1_dims, input1_data, &input1);
+
+  NameMLValMap feeds{{"Concat_output_0", input0}, {"Abs_output_0", input1}};
+
+  std::vector<std::string> output_names{"DequantizeLinear_2_output_0"};
+  std::vector<OrtValue> fetches_orig;
+  std::vector<OrtValue> fetches;
+
+  SessionOptions so;
+  so.session_logid = "QDQTransformerTests.RegressionTest_GitHubIssue21319";
+
+  {
+    so.graph_optimization_level = TransformerLevel::Level1;
+    InferenceSession session{so, GetEnvironment()};
+    ASSERT_STATUS_OK(session.Load(model_uri));
+    ASSERT_STATUS_OK(session.Initialize());
+    ASSERT_STATUS_OK(session.Run(feeds, output_names, &fetches_orig));
+  }
+
+  {
+    so.graph_optimization_level = TransformerLevel::Level2;  // enable qdq transformer
+    InferenceSession session{so, GetEnvironment()};
+    ASSERT_STATUS_OK(session.Load(model_uri));
+    ASSERT_STATUS_OK(session.Initialize());
+    ASSERT_STATUS_OK(session.Run(feeds, output_names, &fetches));
+  }
+
+  ASSERT_THAT(fetches_orig[0].Get<Tensor>().DataAsSpan<float>(),
+              testing::ContainerEq(fetches[0].Get<Tensor>().DataAsSpan<float>()));
+}
+
 TEST(QDQTransformerTests, Resize_No_Fusion) {
   auto test_case = [&](const std::vector<int64_t>& input_shape,
                        const std::vector<int64_t>& sizes_shape,
diff --git a/onnxruntime/test/providers/cpu/tensor/upsample_op_test.cc b/onnxruntime/test/providers/cpu/tensor/upsample_op_test.cc
index 3ac8053aef95e..a398b86fc243d 100644
--- a/onnxruntime/test/providers/cpu/tensor/upsample_op_test.cc
+++ b/onnxruntime/test/providers/cpu/tensor/upsample_op_test.cc
@@ -841,15 +841,15 @@ TEST(UpsampleOpTest, UpsampleOp4DBilinearTest_int32) {
   test.AddInput<int32_t>("X", {N, C, H, W}, X);
 
   std::vector<int32_t> Y = {
-      1, 1, 2, 2, 3, 3, 3, 3,
-      2, 2, 3, 3, 4, 4, 4, 4,
-      3, 3, 4, 4, 5, 5, 5, 5,
-      3, 3, 4, 4, 5, 5, 5, 5,
-
-      3, 3, 4, 4, 5, 5, 5, 5,
-      5, 5, 6, 6, 7, 7, 7, 7,
-      7, 7, 8, 8, 9, 9, 9, 9,
-      7, 7, 8, 8, 9, 9, 9, 9};
+      1, 2 /* 1.5 */, 2, 2, 3, 3, 3, 3,
+      2, 2, 3, 4 /* 3.5 */, 4, 4, 4, 4,
+      3, 4 /* 3.5 */, 4, 4, 5, 5, 5, 5,
+      3, 4 /* 3.5 */, 4, 4, 5, 5, 5, 5,
+
+      3, 4 /* 3.5 */, 4, 4, 5, 5, 5, 5,
+      5, 6 /* 5.5 */, 6, 6, 7, 7, 7, 7,
+      7, 8 /* 7.5 */, 8, 8, 9, 9, 9, 9,
+      7, 8 /* 7.5 */, 8, 8, 9, 9, 9, 9};
 
   test.AddOutput<int32_t>("Y", {N, C, (int64_t)(H * scales[2]), (int64_t)(W * scales[3])}, Y);
   test.Run();
diff --git a/onnxruntime/test/testdata/qdq_transformer/gh_issue_21319.onnx b/onnxruntime/test/testdata/qdq_transformer/gh_issue_21319.onnx
new file mode 100644
index 0000000000000..f9052dc8d32c6
Binary files /dev/null and b/onnxruntime/test/testdata/qdq_transformer/gh_issue_21319.onnx differ