diff --git a/libyuv_test.gyp b/libyuv_test.gyp
index f67a85269..ab8b91bd6 100644
--- a/libyuv_test.gyp
+++ b/libyuv_test.gyp
@@ -24,7 +24,8 @@
          # sources
          'unit_test/unit_test.cc',
          'unit_test/rotate_test.cc',
-      ], # source
+         'unit_test/scale_test.cc',
+      ],
       'conditions': [
         ['OS=="linux"', {
           'cflags': [
diff --git a/source/scale.cc b/source/scale.cc
index 9cc5e6200..d3b7d3332 100644
--- a/source/scale.cc
+++ b/source/scale.cc
@@ -1213,7 +1213,7 @@ static void ScaleRowDown4Int_SSE2(const uint8* src_ptr, int src_stride,
   asm volatile(
   "pcmpeqb    %%xmm7,%%xmm7\n"
   "psrlw      $0x8,%%xmm7\n"
-  "lea        (%4,%4,2),%4\n"
+  "lea        (%4,%4,2),%3\n"
 "1:"
   "movdqa     (%0),%%xmm0\n"
   "movdqa     0x10(%0),%%xmm1\n"
diff --git a/unit_test/scale_test.cc b/unit_test/scale_test.cc
new file mode 100644
index 000000000..e147d78b0
--- /dev/null
+++ b/unit_test/scale_test.cc
@@ -0,0 +1,159 @@
+/*
+ *  Copyright (c) 2011 The LibYuv project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "libyuv/scale.h"
+#include "unit_test.h"
+#include <stdlib.h>
+#include <time.h>
+
+using namespace libyuv;
+
+#define align_buffer_16(var, size) \
+  uint8 *var; \
+  uint8 *var##_mem; \
+  var##_mem = reinterpret_cast<uint8*>(calloc(size+15, sizeof(uint8))); \
+  var = reinterpret_cast<uint8*> \
+        ((reinterpret_cast<intptr_t>(var##_mem) + 15) & (~0x0f));
+
+#define free_aligned_buffer_16(var) \
+  free(var##_mem);  \
+  var = 0;
+
+TEST_F(libyuvTest, ScaleDownBy4) {
+  int b = 128;
+  int src_width = 1280;
+  int src_height = 720;
+  int src_width_uv = (src_width + 1) >> 1;
+  int src_height_uv = (src_height + 1) >> 1;
+
+  int src_y_plane_size = (src_width + (2 * b)) * (src_height + (2 * b));
+  int src_uv_plane_size = (src_width_uv + (2 * b)) * (src_height_uv + (2 * b));
+
+  int src_stride_y = 2 * b + src_width;
+  int src_stride_uv = 2 * b + src_width_uv;
+
+  align_buffer_16(src_y, src_y_plane_size)
+  align_buffer_16(src_u, src_uv_plane_size)
+  align_buffer_16(src_v, src_uv_plane_size)
+
+  int dst_width = src_width >> 2;
+  int dst_height = src_height >> 2;
+
+  int dst_width_uv = (dst_width + 1) >> 1;
+  int dst_height_uv = (dst_height + 1) >> 1;
+
+  int dst_y_plane_size = (dst_width + (2 * b)) * (dst_height + (2 * b));
+  int dst_uv_plane_size = (dst_width_uv + (2 * b)) * (dst_height_uv + (2 * b));
+
+  int dst_stride_y = 2 * b + dst_width;
+  int dst_stride_uv = 2 * b + dst_width_uv;
+
+  align_buffer_16(dst_y, dst_y_plane_size)
+  align_buffer_16(dst_u, dst_uv_plane_size)
+  align_buffer_16(dst_v, dst_uv_plane_size)
+
+  // create an image with random data reoccurring in 4x4 grid.  When the image
+  // is filtered all the values should be the same.
+  srandom(time(NULL));
+
+  uint8 block_data[16];
+
+  int i, j;
+
+  // Pulling 16 random numbers there is an infinitesimally small
+  //  chance that they are all 0.  Then the output will be all 0.
+  //  Output buffer is filled with 0, want to make sure that after the
+  //  filtering something went into the output buffer.
+  //  Avoid this by setting one of the values to 128.  Also set the
+  //  random data to at least 1 for when point sampling to prevent
+  //  output all being 0.
+  block_data[0] = 128;
+
+  for (i = 1; i < 16; i++)
+    block_data[i] = (random() & 0xfe) + 1;
+
+  for (i = b; i < (src_height + b); i += 4) {
+    for (j = b; j < (src_width + b); j += 4) {
+      uint8 *ptr = src_y + (i * src_stride_y) + j;
+      int k, l;
+      for (k = 0; k < 4; ++k)
+        for (l = 0; l < 4; ++l)
+          ptr[k + src_stride_y * l] = block_data[k + 4 * l];
+    }
+  }
+
+  for (i = 1; i < 16; i++)
+    block_data[i] = (random() & 0xfe) + 1;
+
+  for (i = b; i < (src_height_uv + b); i += 4) {
+    for (j = b; j < (src_width_uv + b); j += 4) {
+      uint8 *ptru = src_u + (i * src_stride_uv) + j;
+      uint8 *ptrv = src_v + (i * src_stride_uv) + j;
+      int k, l;
+      for (k = 0; k < 4; ++k)
+        for (l = 0; l < 4; ++l) {
+          ptru[k + src_stride_uv * l] = block_data[k + 4 * l];
+          ptrv[k + src_stride_uv * l] = block_data[k + 4 * l];
+        }
+    }
+  }
+
+  int f;
+  int err = 0;
+
+  // currently three filter modes, defined as FilterMode in scale.h
+  for (f = 0; f < 3; ++f) {
+    I420Scale(src_y + (src_stride_y * b) + b, src_stride_y,
+              src_u + (src_stride_uv * b) + b, src_stride_uv,
+              src_v + (src_stride_uv * b) + b, src_stride_uv,
+              src_width, src_height,
+              dst_y + (dst_stride_y * b) + b, dst_stride_y,
+              dst_u + (dst_stride_uv * b) + b, dst_stride_uv,
+              dst_v + (dst_stride_uv * b) + b, dst_stride_uv,
+              dst_width, dst_height,
+              static_cast<FilterMode>(f));
+
+    int value = dst_y[(dst_stride_y * b) + b];
+
+    // catch the case that the output buffer is all 0
+    if (value == 0)
+      ++err;
+
+    for (i = b; i < (dst_height + b); ++i) {
+      for (j = b; j < (dst_width + b); ++j) {
+        if (value != dst_y[(i * dst_stride_y) + j])
+          ++err;
+      }
+    }
+
+    value = dst_u[(dst_stride_uv * b) + b];
+
+    if (value == 0)
+      ++err;
+
+    for (i = b; i < (dst_height_uv + b); ++i) {
+      for (j = b; j < (dst_width_uv + b); ++j) {
+        if (value != dst_u[(i * dst_stride_uv) + j])
+          ++err;
+        if (value != dst_v[(i * dst_stride_uv) + j])
+          ++err;
+      }
+    }
+  }
+
+  free_aligned_buffer_16(src_y)
+  free_aligned_buffer_16(src_u)
+  free_aligned_buffer_16(src_v)
+  free_aligned_buffer_16(dst_y)
+  free_aligned_buffer_16(dst_u)
+  free_aligned_buffer_16(dst_v)
+
+  EXPECT_EQ(0, err);
+}