Since their introduction, the interaction trap and other two-hybrid systems have been used to study protein-protein interactions. Despite their general use, little is known about the extent to which the degree of protein interaction determined by two-hybrid approaches parallels the degree of interaction determined by biochemical techniques. In this study, we used a set of lexAop-LEU2 and lexAop-lacZ reporters to calibrate the interaction trap. For the calibration, we used two sets of proteins, the Myc-Max-Mxi1 helix-loop-helix proteins, and wild-type and dimerization-defective versions of the lambda cI repressor. Our results indicate that the strength of interaction as predicted by the two-hybrid approach generally correlates with that determined in vitro, permitting discrimination of high-, intermediate-, and low-affinity interactions, but there was no single reporter for which the amount of gene expression linearly reflected affinity measured in vitro. However, some reporters showed thresholds and only responded to stronger interactions. Finally, some interactions were subject to directionality, and their apparent strength depended on the reporter used. Taken together, our results provide a cautionary framework for interpreting affinities from two-hybrid experiments.