A multiproduct cost function is estimated for German higher education institutions using a panel of data from recent years. The use of panel data allows a random parameter stochastic frontier model to be estimated, and this delivers new insights on the extent to which differences in costs between institutions producing similar vectors of outputs may be due to different cost structures, on the one hand, and efficiency, on the other. The approach used here therefore resembles in some respects the non-parametric methods of efficiency evaluation, since different loss functions attach to different universities. We report also on measures of average incremental cost of provision and on returns to scale and scope. (Contains 6 notes and 5 tables.)