This paper considers the use of instruments to identify and estimate private and social returns to education within a model of employer learning. What an instrument identifies depends on whether it is hidden from, or transparent (i.e., observed) to, the employers. A hidden instrument identifies private returns to education, and a transparent instrument identifies social returns to education. We use variation in compulsory schooling laws across non-central and central municipalities in Norway to, respectively, construct hidden and transparent instruments. We estimate a private return of 7.9%, of which 70% is due to increased productivity and the remaining 30% is due to signaling.