diff --git a/paddle/fluid/operators/yolov3_loss_op.h b/paddle/fluid/operators/yolov3_loss_op.h
index de01a01a4fb859e5bd3c5213f86290e6071629f0..2131289860eb9f59253aaf218451b466fcebb3ba 100644
--- a/paddle/fluid/operators/yolov3_loss_op.h
+++ b/paddle/fluid/operators/yolov3_loss_op.h
@@ -41,6 +41,11 @@ static T L1Loss(T x, T y) {
   return std::abs(y - x);
 }
 
+template <typename T>
+static T L2Loss(T x, T y) {
+  return 0.5 * (y - x) * (y - x);
+}
+
 template <typename T>
 static T SCEGrad(T x, T label) {
   return 1.0 / (1.0 + std::exp(-x)) - label;
@@ -51,6 +56,11 @@ static T L1LossGrad(T x, T y) {
   return x > y ? 1.0 : -1.0;
 }
 
+template <typename T>
+static T L2LossGrad(T x, T y) {
+  return x - y;
+}
+
 static int GetMaskIndex(std::vector<int> mask, int val) {
   for (size_t i = 0; i < mask.size(); i++) {
     if (mask[i] == val) {
@@ -130,8 +140,8 @@ static void CalcBoxLocationLoss(T* loss, const T* input, Box<T> gt,
   T scale = (2.0 - gt.w * gt.h) * score;
   loss[0] += SCE<T>(input[box_idx], tx) * scale;
   loss[0] += SCE<T>(input[box_idx + stride], ty) * scale;
-  loss[0] += L1Loss<T>(input[box_idx + 2 * stride], tw) * scale;
-  loss[0] += L1Loss<T>(input[box_idx + 3 * stride], th) * scale;
+  loss[0] += L2Loss<T>(input[box_idx + 2 * stride], tw) * scale;
+  loss[0] += L2Loss<T>(input[box_idx + 3 * stride], th) * scale;
 }
 
 template <typename T>
@@ -150,9 +160,9 @@ static void CalcBoxLocationLossGrad(T* input_grad, const T loss, const T* input,
   input_grad[box_idx + stride] =
       SCEGrad<T>(input[box_idx + stride], ty) * scale * loss;
   input_grad[box_idx + 2 * stride] =
-      L1LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss;
+      L2LossGrad<T>(input[box_idx + 2 * stride], tw) * scale * loss;
   input_grad[box_idx + 3 * stride] =
-      L1LossGrad<T>(input[box_idx + 3 * stride], th) * scale * loss;
+      L2LossGrad<T>(input[box_idx + 3 * stride], th) * scale * loss;
 }
 
 template <typename T>
diff --git a/python/paddle/fluid/tests/unittests/test_yolov3_loss_op.py b/python/paddle/fluid/tests/unittests/test_yolov3_loss_op.py
index ff76b763663ab598758ab52d7ef64460271919d1..0e17eb31305839e7a0ca02c4332956a0b256ed50 100644
--- a/python/paddle/fluid/tests/unittests/test_yolov3_loss_op.py
+++ b/python/paddle/fluid/tests/unittests/test_yolov3_loss_op.py
@@ -27,6 +27,10 @@ def l1loss(x, y):
     return abs(x - y)
 
 
+def l2loss(x, y):
+    return 0.5 * (y - x) * (y - x)
+
+
 def sce(x, label):
     sigmoid_x = expit(x)
     term1 = label * np.log(sigmoid_x)
@@ -145,8 +149,8 @@ def YOLOv3Loss(x, gtbox, gtlabel, gtscore, attrs):
             scale = (2.0 - gtbox[i, j, 2] * gtbox[i, j, 3]) * gtscore[i, j]
             loss[i] += sce(x[i, an_idx, gj, gi, 0], tx) * scale
             loss[i] += sce(x[i, an_idx, gj, gi, 1], ty) * scale
-            loss[i] += l1loss(x[i, an_idx, gj, gi, 2], tw) * scale
-            loss[i] += l1loss(x[i, an_idx, gj, gi, 3], th) * scale
+            loss[i] += l2loss(x[i, an_idx, gj, gi, 2], tw) * scale
+            loss[i] += l2loss(x[i, an_idx, gj, gi, 3], th) * scale
 
             objness[i, an_idx * h * w + gj * w + gi] = gtscore[i, j]
 
@@ -202,7 +206,7 @@ class TestYolov3LossOp(OpTest):
 
     def test_check_output(self):
         place = core.CPUPlace()
-        self.check_output_with_place(place, atol=2e-3)
+        self.check_output_with_place(place, atol=1e-3)
 
     def test_check_grad_ignore_gtbox(self):
         place = core.CPUPlace()
@@ -210,19 +214,16 @@ class TestYolov3LossOp(OpTest):
             place, ['X'],
             'Loss',
             no_grad_set=set(["GTBox", "GTLabel", "GTScore"]),
-            max_relative_error=0.2)
+            max_relative_error=0.3)
 
     def initTestCase(self):
-        self.anchors = [
-            10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198,
-            373, 326
-        ]
-        self.anchor_mask = [0, 1, 2]
-        self.class_num = 10
-        self.ignore_thresh = 0.7
+        self.anchors = [10, 13, 16, 30, 33, 23]
+        self.anchor_mask = [1, 2]
+        self.class_num = 5
+        self.ignore_thresh = 0.5
         self.downsample = 32
         self.x_shape = (3, len(self.anchor_mask) * (5 + self.class_num), 5, 5)
-        self.gtbox_shape = (3, 10, 4)
+        self.gtbox_shape = (3, 5, 4)
         self.use_label_smooth = True