This paper studies the optimal transmission design for multiple antenna multiple access channels with linear MMSE receiver at the base station and partial channel state information at the mobiles. The performance criterium for optimization is based on a Schur-concave function that works on the average individual MSEs of all users. The optimal beamforming matrix of each user corresponds to the eigenvector matrix of his channel correlation matrix. The remaining power allocation problem can be solved at reduced complexity. It is also direct to extend the results to Schur-convex functions. The results are illustrated by numerical simulations.