This paper indicates that for computer simulations to be seen as a legitimate methodology in social science, and for new knowledge to be generated, serious consideration needs to be given to how simulations could or should be replicated. It is noted that computer simulation models have changed the ways in which researchers are able to observe and study social phenomena such as crime, and that the ability of researchers to replicate the work of others is fundamental to a cumulative science, yet this is noted to rarely occur in computer simulations. The work develops the concept of “systematic replication,” defining it as a method for developing simulation experiments that move towards a generalizable inference which is directed, explicit, and incorporates complexity incrementally. As part of a detailed case study provided, the work puts forth five criteria for systematic replication: testing for presupposed emergence as a rival hypothesis; generalizing about experimental units; generalizing about experimental treatments; generalizing about experimental outcomes; and generalizing about experimental settings. The authors outline how the discrete parts of this process might be carried out in practice, using a simple simulation model. Tables, figure, references