The article discusses agent-based simulation as a tool of sociological understanding. Based on aninferential account of understanding, it argues that computer simulations increase ourexplanatory understanding both by expanding our ability to make what-if inferences about socialprocesses and by making these inferences more reliable. However, our ability to understandsimulations limits our ability to understand real world phenomena through them. ThomasSchelling’s checkerboard model of ethnic segregation is used to demonstrate the important roleplayed by abstract how-possibly models in the process of building a mechanistic understandingof social phenomena.