Cys2His2 zinc finger proteins are composed of modular DNA-binding domains and provide an excellent framework for the design and selection of proteins with novel site specificity. Crystal structures of zinc finger-DNA complexes have shown that many Cys2His2 zinc fingers use a conserved docking arrangement that juxtaposes residues at key positions in the "recognition helix" with corresponding base positions in the three to four base-pair subsite. Several groups have proposed that specificity can be explained with a zinc finger-DNA recognition code that correlates specific amino acids at these key positions in the alpha-helix with specific bases in each position of the corresponding subsite. Here, we explore the utility of such a code through detailed studies of zinc finger variants selected via phage display. These proteins provide interesting systems for detailed analysis since they have affinities and specificities for their sites similar to those of naturally occurring DNA-binding proteins. Comparisons are facilitated by the fact that only key DNA-binding residues are varied in each finger while leaving all other regions of the structure unchanged. We study these proteins in detail by (1) selecting their optimal binding sites and comparing these binding sites with sites that might have been predicted from a code; (2) by examining the "evolutionary history" of these proteins during the phage display protocol to look for evidence of context-dependent effects; and (3) by reselecting finger 1 in the presence of the optimized finger 2/finger 3 domains to obtain further data on finger modularity. Our data for optimized fingers and binding sites demonstrate a clear correlation with contacts that would be predicted from a code. However, there are enough examples of context-dependent effects (not explained by any existing code) that selection is the most reliable method for maximizing the affinity and specificity of new zinc finger proteins.