In this paper we consider Zellner's estimator in the two SUR equations model used in Revankar (1974, 1976) when the disturbances are distributed as a matric variate compound normal distribution which includes a matric t distribution. We show the risks of the unrestricted and the restricted Zellner's estimators and derive the region where the restricted Zellner's estimator dominates the unrestricted one. As a result, we show that as the distribution of disturbances becomes far from the normal distribution, the region becomes small.