Abstract The usual joint optimization of stocking levels and rotation length discussed under the deterministic assumption is generalized to probabilistic cases. A probabilistic dynamic programming model is constructed to resolve the difficulties introduced. Every value in the deterministic model is replaced by an expected value in the probabilistic model, and the objective is to optimize the expected value. An example showing how to use the developed model to solve the optimal stocking levels and rotation problems under probabilistic growth for Douglas-fir is presented. The effects of various degrees of indeterminateness, or risk in growth prediction, are that for larger variance of growth prediction, optimal regimes involve shorter rotation, lower stocking levels and lower mean annual increment. Forest Sci. 28:711-719.