We use a combination of methods - sequence analysis, multiple structure alignment, calculation of folding and binding free energies, and surface property analysis - as a basis for the prediction of protein structure and function. Recent developments in protein structure prediction to be discussed include structure-based sequence alignment methods, new approaches to loop and side chain prediction, and a simple physical-chemical based scoring function. Our approach to the prediction of function from structure involves the recognition of features on protein surfaces that are involved in binding to other proteins, nucleic acids, and membranes. In this context, we have developed an approach to calculate the binding free energies associated with the formation of protein complexes and to identify the contribution of individual amino acids to binding. When this method is combined with structure-based alignment tools, we are able to cluster protein families into functional subgroups and to detect novel sequence/structure/function relationships. Applications to SH2 domains and to C2 domains will be described. In this context, homology modeling is shown to be an effective means of extending and generalizing observations derived from the analysis of known structures.