This paper shows that conventional empirical estimates of matching functions are systematically biased, because unobservable and endogenous search behaviour, such as on-the-job search by workers or the use of different search channels by firms, is neglected. I propose an approach for recovering...