Shapelets are short, interpretable patterns in temporal data which can be characteristic of a class. In this paper, we identify shapelets from the trajectories of discrete-event simulation to indicate the characteristic dynamic behaviours of competing system alternatives. This deviates from traditional simulation output analysis, in which estimations of time-averaged performance measures overlook the more fine-grained time-dynamic features that shape the evolution of a system. We propose a shapelet methodology tailored towards simulation trajectories, and provide mathematical observations to support its implementation. To illustrate the potential of this methodology, we demonstrate its application to three examples. In particular, we reveal disruption recovery behaviour in a manufacturing simulation, provide a means for dynamic model validation, and expose the typical joint behaviour of a multivariate system state. By offering a visual characterisation of trajectories, we find that simulation shapelets can promote a deeper understanding of the dynamic behaviour and performance of simulated models.