This paper describes some simple simulation models of vocabulary attrition. The attrition process is modelled using a random autonomous Boolean network model, and some parallels with real attrition data are drawn. The paper argues that applying a complex systems approach to attrition can provide some important insights, which suggest that real attrition data may need to be treated with caution. It concludes that simulation methods--a technique that is not widely used in applied linguistic research--may be able to throw interesting new light on attrition, and provide a way of examining some of the assumptions that we take for granted in attrition experiments with human subjects.