The task of assigning function to the ever-increasing numbers of genes arising from the genome projects has prompted efforts to exploit structural information to infer biological activity. The conservation of protein folds suggests that proteins can be categorized into families on the basis of sequence similarity. Using this approach, unknown proteins with a fold similar to another protein of known activity can be assumed to have a similar function. Present estimates suggest around 5,000–15,000 new protein structures would provide enough information to model almost all possible proteins. In the review on p. 283, Skolnick et al. describe pilot structural genomics projects currently under way, theoretical approaches for predicting structure/function from sequence, and the biological implications of using structural information.