Models of fishing behavior rarely incorporate the complexities of marine ecosystems, multiple-stock harvest technologies, and regulations present in real world marine fisheries. We introduce a structural model of a multi-species, weak-output-disposability harvest technology. A latent target-cost-minimizing share vector is estimated to link the technology to a spatially and temporally heterogeneous fish stock. Data from the Gulf of Mexico reef fish fishery is used to estimate the model. The results provide a robust characterization of harvest and discard behavior across space and time. Our approach considerably improves methods used to study fishing behavior and evaluate alternative fisheries management policies.