Agent-based models (ABMs) are an increasingly popular tool in the social sciences. This trend seems likely to continue, so that they will become widely used in geography and in urban and regional planning. We present an overview of examples of these models in the life sciences, economics, planning, sociology, and archaeology. We conclude that ABMs strongly tend towards an individualist view of the social world. This point is reinforced by closer consideration of particular examples. This discussion pays attention to the inadequacy of an individualist model of society with reference to debates in social theory. We argue that because models are closed representations of an open world it is important that institutions and other social structures be explicitly included, or that their omission be explained. A tentative explanation for the bias of ABMs is offered, based on an examination of early research in artificial intelligence and distributed artificial intelligence from which disciplines the approach is derived. Some implications of these findings are discussed. We indicate some useful research directions which are beginning to tackle the individualism issue directly. We further note that the underlying assumptions of ABMs are often hidden in the implementation details. We conclude that such models must be subject to critical examination of their assumptions, and that model builders should engage with social theory if the approach is to realise its full potential.