We constructed a simulation model for Redfish Lake, Idaho, using water budget and nutrient loading measurements, to predict the dependence of lake production on nutrients from the watershed, precipitation, lake fertilization, and marine-derived nutrients from sockeye salmon Oncorhynchus nerka, which historically have reared in the lake. We also used the model to simulate different management scenarios to help restore the endangered Snake River sockeye salmon. The model and other empirical evidence indicated that even before hydropower dams were present in the migration corridor, marine-derived nutrients were not of major importance to lake production, contributing only about 3% of the annual phosphorus load of the lake. This contribution was partially removed by the quick flushing rate (3 years) of the lake and phosphorus export by smolts. The model predicted annual adult sockeye salmon returns to be 3,800 fish under predam conditions, 370 fish under modern conditions, 780 fish when watershed nutrient loading was doubled (simulating lake fertilization), and 750 fish when smolt-to-adult survival was doubled. Although fertilization should stimulate sockeye salmon production, the effect would be transitory. The model predicted that 8 years after the end of a 3-year fertilization period, adult returns would be only 5% greater than those for unfertilized conditions. Our analysis suggests that to restore self-sustaining anadromous sockeye salmon populations to Redfish Lake, increased smolt-to-adult survival must be achieved; however, lake fertilization should be considered an important short-term tool for decreasing continued erosion of the stock.