We have developed a substrate-phage approach for examining the substrate specificities of an important group of proteases involved in apoptosis--the caspases. After establishing selection conditions with caspases-3 and caspase-8 vs control substrate-phage, we sorted X4 and X6 diversity libraries, identified consensus motifs that agree with previously defined caspase substrate motifs, confirmed the selection of active substrates using synthetic peptide rate assays under a range of buffer conditions, and compared kinetic parameters for selected substrates. The libraries produced some variations on the canonical motifs. From caspase-3 selections, a phage-derived synthetic peptide, DLVD, was hydrolyzed up to 170% faster than the canonical substrate DEVD. The P4 Asp residue was essential for good protease-sensitivity, but even substrates with substitutions at P4 were selected by phage and shown to be hydrolyzed. Caspase-8 selections, as expected, yielded predominantly clones containing a Glu at P3. In this case, the most frequent phage-derived peptide, LEVD, was cleaved at a rate of only 20% of the canonical caspase-8 substrate LETD. However, based on substitutions observed in the phage selectants at P4, a substrate peptide, AETD, was designed and shown to be hydrolyzed up to 160% faster than LETD. We consider factors that may contribute to differences in caspase substrate-phage selections vs synthetic peptide studies on the caspases, and suggest that the two approaches may offer complementary information.