Education may enhance earnings either because of human capital increases or by signalling unobservable worker attributes. Previous tests of these alternatives relied on ad hoc distinctions between them. Our theoretical model provides a direct signal measure as the difference between required and necessary qualifications, forming the basis for an ordered-probit model of discrete differences between required and necessary qualifications that depend on firm, job, and worker attributes. Estimates support the model's predictions, providing a continuous, predicted signal measure for second-stage earnings equations. These results provide the first formal evidence of a significant, positive, gender-specific return to a signal and indicate a downward bias in the return to education from excluding the signal measure.