Many biological processes rely on protein-protein interactions. These processes include signal transduction, cell cycle regulation, gene regulation, and viral assembly and replication. Moreover, many proteins and enzymes manifest their function as oligomers. We describe here an efficient means to sift through large combinatorial libraries and identify molecules that block the interaction of target proteins in vivo. The power of this approach is demonstrated by the identification of nine-residue peptides from a combinatorial library that inhibit the intracellular dimerization of HIV-1 protease. Fewer than 1 in 106 peptides do so. In vitro biochemical analyses of one such peptide demonstrate that it acts by dissociating HIV-1 protease into monomers, which are inactive catalysts. Inhibition is enhanced further by dimerizing the peptide. This approach enables the facile identification of new molecules that control cellular processes.