diff --git a/tensorflow/lite/kernels/internal/optimized/sse_check.h b/tensorflow/lite/kernels/internal/optimized/sse_check.h
index 9804afaa571..7eb3677f271 100644
--- a/tensorflow/lite/kernels/internal/optimized/sse_check.h
+++ b/tensorflow/lite/kernels/internal/optimized/sse_check.h
@@ -15,14 +15,20 @@ limitations under the License.
 #ifndef TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_SSE_CHECK_H_
 #define TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_SSE_CHECK_H_
 
-#if defined(__SSSE3__)
-// SSSE 3 available: Use the SSE code.
+#if defined(__SSE4_1__)
+// SSE 4.1 available: Use the SSE code.
 #define SSE_OR_PORTABLE(funcname, ...) Sse##funcname(__VA_ARGS__)
 
 #else
 
-// No SSSE 3 available: Use Portable code
-#define SSE_OR_PORTABLE(...) Portable##funcname(__VA_ARGS__)
+#include "tensorflow/lite/kernels/internal/optimized/neon_check.h"
+
+// No SSE 4.1 available: Fall back to NEON_OR_PORTABLE, potentially used with
+// NEON_2_SSE translator library. As the library requires SSSE3, the fallback is
+// generally using Portable code, only a narrow subset of processors supporting
+// SSSE3 but no SSE4.1 is affected - but that includes the android_x86 ABI (not
+// android_x86_64).
+#define SSE_OR_PORTABLE(...) NEON_OR_PORTABLE(__VA_ARGS__)
 #endif
 
 #endif  // TENSORFLOW_LITE_KERNELS_INTERNAL_OPTIMIZED_SSE_CHECK_H_