We study a farmer’s decision to convert traditional crop land into growing dedicated energy crops, taking in account sunk conversion costs, uncertainties in traditional and energy crop returns, and learning. The optimal decision rules differ significantly from the expected net present value rule, which ignores learning, and from real option models that allow only one way conversions into energy crops. These models also predict drastically different patterns of land conversions into and out of energy crops over time. Using corn-soybean rotations and switchgrass as examples, we show that the model predictions are sensitive to assumptions about stochastic processes of the returns. Government policies might have unintended consequences: subsidizing conversion costs into switchgrass reduces proportions of land in switchgrass in the long run.