Using field-experimental data (study-time tracking and randomized incentives), we identify a structural model of learning. Student effort is influenced by external costs/benefits and unobserved heterogeneity: motivation (willingness to study) and productivity (conversion rate of time into skill). We estimate academic labor-supply elasticities and skill technology. Productivity and motivation are uncorrelated. Low productivity, not low motivation, is the stronger predictor of academic struggles. School quality augments productivity and accelerates skill production. We find that dynamic skill complementarities arise mainly from children's aging and from a feedback loop between investment activity and productivity, rather than from carrying forward past skill stocks.