Summary: In this paper, the authors investigate the empirical Bayes estimation of parameters and its superiority in multivariate linear models with respect to normal-inverse Wishart priors. When the parameters of prior distribution are partly unknown, the empirical Bayes estimators of the regression coefficient matrix and the error variance matrix are constructed. It is shown that the empirical Bayes estimators are superior to the corresponding least square estimators under the criteria of Bayes mean square error (BMSE for short) and Bayes mean square error matrix (BMSEM for short). Finally, a Monte Carlo simulation is carried out to verify the theoretical results.