Abstract: Multi-classification method under t distribution is proposed in order to solve the problem of the traditional face classification methods failing to settle tailing situation with heavy-tailed noise. By adjusting the degree of freedom parameter v, t distribution will adapt to a variety of tailing conditions better after heavy-tailed noise is added in facial images, at the same time, the recognition results will be improved. The experimental results on ORL and Yale show the feasibility and the effectiveness of the proposed algorithm.