In vitro selection from molecular libraries has rapidly come of age as a protein-engineering tool. Dramatic increases in protein affinity can be engineered using phage-display libraries, and specific antibodies can be selected directly from a single 'naïve' library of their genes. Repertoires of small molecules are a potentially valuable resource for drug discovery. Libraries of linear peptides provide ligands for proteins that recognize continuous epitopes, and low-affinity mimics of some small molecules, but generally do not contain mimics of large molecular interfaces. Switching to constrained peptide formats, and deploying more diverse, non-peptide chemical libraries, may bring greater success.