Large-scale sequencing projects have predicted high numbers of gene products for which no functional information is yet available. Hence, large-scale projects, such as gene knockouts, gene expression profiles, and protein-interaction mapping, are currently under way to initiate the understanding of the function of these gene products. The high-throughput strategies that are currently being developed to generate protein-interaction maps include automated versions of the yeast two-hybrid system. These strategies rely on the large-scale construction of DNA-binding domain/protein-of-interest hybrid constructs (DB-X baits). An inherent problem of large-scale two-hybrid systems is that a high percentage of cloned sequences encode polypeptides that, when fused to DB, can activate transcription in the absence of any two-hybrid-interacting partner protein. Here, we describe and validate a genetic strategy that efficiently eliminates such self-activator baits prior to screening procedures. The strategy is based on a negative-growth selection and is compatible with high-throughput settings.