A technique was developed for studying protein-DNA recognition that can be applied to any purified protein, partially purified protein, or cloned gene. From oligonucleotides in which particular positions are of random sequence, that subset to which a given protein binds is amplified by the polymerase chain reaction and sequenced as a pool. These selected and amplified binding site (SAAB) "imprints" provide a characteristic set of preferred sequences for protein binding. With this technique, it was shown that homo- and heterooligomers of the helix-loop-helix proteins MyoD and E2A recognize a common consensus sequence, CA--TG, but otherwise bind to flanking and internal positions with different sequence preferences that suggest half-site recognition. These findings suggest that different combinations of dimeric proteins can have different binding sequence preferences.