Autism constitutes one of the best validated child psychiatric disorders. Empirical research has succeeded in delineating the key clinical phenomena, in demonstrating strong genetic influences on the underlying liability, and in identifying basic cognitive deficits. A range of neurobiological abnormalities has also been found, although the replicability of specific findings has not been high. An understanding of the causal processes leading to autism, and accounting for the marked variability in its manifestations, requires an integration across these different levels of enquiry. Although this is not yet possible, a partial integration provides a useful strategy for identifying key research questions, the limitations of existing hypotheses, and future research directions that are likely to prove fruitful. The research findings for each research level are critically reviewed in order to consider how to move towards an integration across levels.