New ligands for a variety of biological targets can be selected from biological or synthetic combinatorial peptide libraries. The use of different libraries to select novel peptides with potential therapeutic applications is reviewed. The possible combination of molecular diversity provided by combinatorial libraries and a rational approach derived from computational modeling is also considered. Advantages and disadvantages of different approaches are compared. Possible strategies to bypass loss of peptide bioactivity in the transition from ligand selection to in vivo use are discussed.