Abstarct: After magnificent work of Prof Charles Stein in 1956, regarding the inadmissibility of the minimax estimator of mean of multivariate normal, when the dimension is greater than 2, there are many studies to develop improved estimators in the sense of having smaller risk or MSE that the minimax estimator of mean under different settings. These improved estimators are shrinkage in some sense. In this approach, we consider multivariate models with matrix structure for errors. We study the performance of least squares as well as shrinkage estimators. The underlying models are matrix variate normal and matrix variate elliptical distributions. The risk functions of the proposed estimators are exactly derived and the performance of the estimators is compared by their risk measures.