diff --git a/tensorflow/lite/kernels/detection_postprocess.cc b/tensorflow/lite/kernels/detection_postprocess.cc
index a0df4a10fa1bb2..09ed7945ec574b 100644
--- a/tensorflow/lite/kernels/detection_postprocess.cc
+++ b/tensorflow/lite/kernels/detection_postprocess.cc
@@ -130,7 +130,7 @@ TfLiteStatus SetTensorSizes(TfLiteContext* context, TfLiteTensor* tensor,
                             std::initializer_list<int> values) {
   TfLiteIntArray* size = TfLiteIntArrayCreate(values.size());
   int index = 0;
-  for (int v : values) {
+  for (const auto& v : values) {
     size->data[index] = v;
     ++index;
   }
@@ -503,7 +503,7 @@ TfLiteStatus NonMaxSuppressionMultiClassRegularHelper(TfLiteContext* context,
         num_detections_per_class));
     // Add selected indices from non-max suppression of boxes in this class
     int output_index = size_of_sorted_indices;
-    for (int selected_index : selected) {
+    for (const auto& selected_index : selected) {
       box_indices_after_regular_non_max_suppression[output_index] =
           (selected_index * num_classes_with_background + col + label_offset);
       scores_after_regular_non_max_suppression[output_index] =
diff --git a/tensorflow/lite/kernels/kernel_util_test.cc b/tensorflow/lite/kernels/kernel_util_test.cc
index 4e792542a19eaf..2f581feb1ba4fd 100644
--- a/tensorflow/lite/kernels/kernel_util_test.cc
+++ b/tensorflow/lite/kernels/kernel_util_test.cc
@@ -44,7 +44,7 @@ class KernelUtilTest : public ::testing::Test {
     TfLiteTensorFree(tensor);
     tensor->dims = TfLiteIntArrayCreate(dims.size());
     int i = 0;
-    for (int d : dims) {
+    for (const auto& d : dims) {
       tensor->dims->data[i] = d;
       ++i;
     }
diff --git a/tensorflow/lite/kernels/test_util.h b/tensorflow/lite/kernels/test_util.h
index 08c027f9d9d4fb..a4207a5e8179de 100644
--- a/tensorflow/lite/kernels/test_util.h
+++ b/tensorflow/lite/kernels/test_util.h
@@ -46,7 +46,7 @@ template <typename T>
 inline std::vector<T> Quantize(const std::vector<float>& data, float scale,
                                int32_t zero_point) {
   std::vector<T> q;
-  for (float f : data) {
+  for (const auto& f : data) {
     q.push_back(static_cast<T>(std::max<float>(
         std::numeric_limits<T>::min(),
         std::min<float>(std::numeric_limits<T>::max(),
@@ -59,7 +59,7 @@ template <typename T>
 inline std::vector<float> Dequantize(const std::vector<T>& data, float scale,
                                      int32_t zero_point) {
   std::vector<float> f;
-  for (T q : data) {
+  for (const T& q : data) {
     f.push_back(scale * (q - zero_point));
   }
   return f;
@@ -276,7 +276,7 @@ class SingleOpModel {
                << ". Requested " << typeToTfLiteType<T>() << ", got "
                << t->type;
     }
-    for (T f : data) {
+    for (const T& f : data) {
       *v = f;
       ++v;
     }
@@ -296,7 +296,7 @@ class SingleOpModel {
                << ". Requested " << typeToTfLiteType<T>() << ", got "
                << t->type;
     }
-    for (T f : data) {
+    for (const T& f : data) {
       *v = f;
       ++v;
     }
@@ -496,7 +496,7 @@ class SingleOpTest : public ::testing::TestWithParam<string> {
   static std::vector<string> GetKernelTags(
       const std::map<string, TfLiteRegistration*>& kernel_map) {
     std::vector<string> tags;
-    for (auto it : kernel_map) {
+    for (const auto& it : kernel_map) {
       tags.push_back(it.first);
     }
     return tags;
diff --git a/tensorflow/lite/toco/tflite/export_test.cc b/tensorflow/lite/toco/tflite/export_test.cc
index fb640f776abdef..7d5b59ee661bbf 100644
--- a/tensorflow/lite/toco/tflite/export_test.cc
+++ b/tensorflow/lite/toco/tflite/export_test.cc
@@ -251,7 +251,7 @@ class OpSetsTest : public ExportTest {
     params_.enable_select_tf_ops = false;
     params_.quantize_weights = false;
 
-    for (OpSet i : sets) {
+    for (const OpSet& i : sets) {
       switch (i) {
         case kTfLiteBuiltins:
           import_all_ops_as_unsupported_ = false;