Video inpainting is to recover the distorted pixels of the image sequences. In this paper, we use Gibbs distribution and Markov Random Field (MRF) modeling to solve this problem. We build MRF on the missing region of the image sequence conditioned on its neighborhood pixels that are undistorted. We recover the missing pixels by drawing samples from the MRF by Gibbs sampling. We collect a dataset to test the method. The experiments on the dataset with different types of missing pixels show that the method works well in both quantitative and qualitative aspects.