In this paper we introduce marginal cost criteria to determine the optimum time for preventively replacing a fixed group of components. The criteria have a clear interpretation and are very flexible: actual ages can be taken into account as well as discretely or continuously distributed lifetimes and any number of components. Since an overall optimal policy for the group replacement problem is difficult to establish, we compare the creteria with block replacement policies which replace the group at fixed intervals. Such policies, however, do not take failure renewals into account and may replace relatively new components. The performance of the criteria has been analysed both with discrete-time Markov decision chains and with simulation. In all cases considered the replacement criteria yield an improvement in average costs over the optimum block replacement policy varying between 0% and 10%, while the loss with respect to an optimal group age replacement policy (if it can be determined at all) is marginal.