In a literature on group replacement policies for batches of machines, it is generally assumed that the machines are identical with known parameters and the objective is to find the policy that minimizes expected cost per unit time. The objective of this research is to investigate ways of relaxing these assumptions. As with all stochastically evolving systems, variance should play an important managerial role. It is demonstrated that it is both feasible and practicable to calculate the variance of the cost per unit time associated with group maintenance policies. For the case of machines with different unit replacement costs, production rates and failure times parameters, an algorithm for reducing the size of the policies' set is presented. It is argued that, when different types of machine are available, the form of replacement policy should influence purchasing decisions. When the failure time distribution is of phase type, explicit results for the expected cost and variance per unit time of policy in use are obtained. For the situation where failure times parameters are not known, optimal Bayesian group replacement policies are presented.