This article explores some critical methodological and theoretical issues that emerge from recent research into word association behaviour in second language (L2) learners. The studies that we discuss here all use computer simulations as a tool to investigate L2 lexical networks, and to compare these networks with those of first language (L1) speakers. This article broaches some previously unacknowledged complexities in this kind of research, and draws attention to the importance of which assumptions are built into simulation models. The article queries some of the assumptions of our 2002 article (Wilks and Meara, 2002), and provides a reinterpretation of some of the data that we presented there. The article argues that simulation modelling forces us to make critical analyses of assumptions in a way that is not always necessary in less exacting experimental environments.