diff --git a/paddle/phi/kernels/impl/matmul_kernel_impl.h b/paddle/phi/kernels/impl/matmul_kernel_impl.h
index f7b86ea8bf4045..d087e25d1e1fa3 100644
--- a/paddle/phi/kernels/impl/matmul_kernel_impl.h
+++ b/paddle/phi/kernels/impl/matmul_kernel_impl.h
@@ -2007,16 +2007,36 @@ void MatmulKernel(const Context& ctx,
                   bool transpose_x,
                   bool transpose_y,
                   DenseTensor* out) {
-  PADDLE_ENFORCE_NE(common::product(x.dims()),
-                    0,
-                    common::errors::InvalidArgument(
-                        "The Input(X) dims size must not be equal "
-                        "0, but received dims size is 0."));
-  PADDLE_ENFORCE_NE(common::product(y.dims()),
-                    0,
-                    common::errors::InvalidArgument(
-                        "The Input(Y) dims size must not be equal "
-                        "0, but received dims size is 0."));
+  if (x.numel() == 0 || y.numel() == 0) {
+    auto x_dims = x.dims();
+    auto y_dims = y.dims();
+    if (transpose_x) {
+      std::swap(x_dims[x_dims.size() - 1], x_dims[x_dims.size() - 2]);
+    }
+    if (transpose_y) {
+      std::swap(y_dims[y_dims.size() - 1], y_dims[y_dims.size() - 2]);
+    }
+    std::vector<std::int64_t> out_dims(x_dims.size() - 1 + y_dims.size() - 1);
+    for (int64_t i = 0; i < x_dims.size() - 1; ++i) {
+      out_dims[i] = x_dims[i];
+    }
+    for (int64_t i = 1; i < y_dims.size(); ++i) {
+      out_dims[x_dims.size() - 1 + i - 1] = y_dims[i];
+    }
+    out->Resize(phi::make_ddim(out_dims));
+    ctx.template Alloc<T>(out);
+    return;
+  }
+  PADDLE_ENFORCE_GE(
+      common::product(x.dims()),
+      0,
+      common::errors::InvalidArgument(
+          "The dims of Input(X) should be greater than or equal to 0."));
+  PADDLE_ENFORCE_GE(
+      common::product(y.dims()),
+      0,
+      common::errors::InvalidArgument(
+          "The dims of Input(Y) should be greater than or equal to 0."));
   const std::vector<std::int64_t> x_dims = common::vectorize(x.dims());
   const std::vector<std::int64_t> y_dims = common::vectorize(y.dims());
   MatmulJudgeDtypeKernel<Context, T>(
diff --git a/test/legacy_test/test_tensordot.py b/test/legacy_test/test_tensordot.py
index 3d15a0f7931b4d..1df90d2a226267 100644
--- a/test/legacy_test/test_tensordot.py
+++ b/test/legacy_test/test_tensordot.py
@@ -371,5 +371,116 @@ def set_dtype(self):
         self.dtype = np.float64
 
 
+class TestTensordotAPIZeroSize(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [0, 5, 5, 5]
+        self.y_shape = [0, 5, 5, 5]
+
+    def set_input_data(self):
+        self.x = np.random.random(self.x_shape).astype(self.dtype)
+        self.y = np.random.random(self.y_shape).astype(self.dtype)
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestTensordotAPIFloat64ZeroSize(TestTensordotAPIZeroSize):
+    def set_dtype(self):
+        self.dtype = np.float64
+
+
+class TestTensordotAPIZeroSize(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [0, 5, 5, 5]
+        self.y_shape = [0, 5, 5, 5]
+
+    def set_input_data(self):
+        self.x = np.random.random(self.x_shape).astype(self.dtype)
+        self.y = np.random.random(self.y_shape).astype(self.dtype)
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+    def set_dtype(self):
+        self.dtype = np.float64
+
+
+class TestTensordotAPIZeroSizeMultipleDims1(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [0, 0, 5, 5]
+        self.y_shape = [0, 0, 5, 5]
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestTensordotAPIZeroSizeMultipleDims2(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [5, 0, 5, 0]
+        self.y_shape = [5, 0, 5, 0]
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestTensordotAPIZeroSizeDifferentDims1(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [5, 5, 0, 5]
+        self.y_shape = [5, 5, 0, 5]
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestTensordotAPIZeroSizeDifferentDims2(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [5, 5, 5, 0]
+        self.y_shape = [5, 5, 5, 0]
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestTensordotAPISingleElementAndZeroSize(TestTensordotAPI):
+    def set_input_shape(self):
+        self.x_shape = [1, 5, 5, 5]
+        self.y_shape = [0, 5, 5, 5]
+
+    def set_test_axes(self):
+        self.all_axes = [
+            [[], []],
+        ]
+
+
+class TestBroadcastWithZeroSize1(unittest.TestCase):
+    def setUp(self):
+        self.x_shape = [5, 0, 3]
+        self.y_shape = [3, 4, 0]
+
+    def set_test_axes(self):
+        self.all_axes = [[], []]
+
+
+class TestBroadcastWithZeroSize2(unittest.TestCase):
+    def setUp(self):
+        self.x_shape = [5, 0, 3]
+        self.y_shape = [3, 0]
+
+    def set_test_axes(self):
+        self.all_axes = [[], []]
+
+
 if __name__ == "__main__":
     unittest.main()