Abstract--This correspondence derives lower bounds on the meansquare error (MSE) for the estimation of a covariance matrix , using samples k = 1; . . . ; K, whose covariance matrices are randomly distributed around . This framework can be encountered e.g., in a radar system operating in a nonhomogeneous environment, when it is desired to estimate the covariance matrix of a range cell under test, using training samples from adjacent cells, and the noise is nonhomogeneous between the cells. We consider two different assumptions for . First, we assume that is a deterministic and unknown matrix, and we derive the Cram