Abstract
Let F p×phave a multivariate F distribution with a scale p×p matrix Δ and degrees of freedom k1 and k2 such that ki - p - 1 > 0, i = 1,2. The estimation of Δ under entropy and squared error loss functions are considered. In both cases a new class of orthogonally invariant estimators are obtained which dominate the best unbiased estimator.