In this paper, an improved BMA(Boundary Matching Algorithm) using gradient vectors is presented to conceal channel errors in inter-frames of H.264 video images. BMA computes the sum of pixel differences around the perimeter of the lost block between the candidate block and its neighboring blocks to estimate the validity of the candidate block, assuming that adjacent pixels in an image have almost the same value. In real images, however, there exist some gradients in local area of the image, which means that the pixel values are increasing or decreasing in a specific slope. A simple and efficient method for estimating candidate blocks using gradient information has been developed and thereby the modified BMA has been applied to recover lost blocks. Experiments show the proposed method improves picture quality of about 1 3dB compared to existing methods.