Abstract We introduce a symmetry reduction technique for model checking temporal-epistemic properties of multi-agent systems defifined in the mainstream interpreted systems framework. The technique, based on counterpart semantics, aims to reduce the set of initial states that need to be considered in a model. We present theoretical results establishing that there are neither false positives nor false negatives in the reduced model. We evaluate the technique by presenting the results of an implementation tested against two well known applications of epistemic logic, the muddy children and the dining cryptographers. The experimental results obtained confifirm that the reduction in model checking time can be dramatic, thereby allowing for the verifification of hitherto intractable systems.