We used stochastic models for analysis of the uncertainty involved in semi-intensive production of shrimp in Nayarit state, Mexico, incorporating partial harvests.Analysis of the database showed that increasing the number of harvests was associated with lower stocking densities, the use of larger ponds, longer cultivation periods, larger final weight of shrimp and total production. Equivalence tests showed that the models adequately fitted the primary data. Monte Carlo simulations indicated that improving management by controlling stocking density and the duration of cultivation increased mean production from 981 to 2573 kg/ha (one partial harvest), from 1808 to 3602 kg/ha (two partial harvests) and from 1364 to 3834 kg/ha (three partial harvests), closely approaching the yields reported in the database. When conducting one and three partial harvests, improved management increased production and the certainty in obtaining the crops, as indicated by diminishing values of the coefficient of variation in output probability distributions.When conducting two partial harvests, however, improved management increased yields, but also increased uncertainty because there was a lower control on production parameters. This does not necessarily imply more uncertainty when conducting two harvests, but that at this stage of knowledge, the primary data only allows detecting limited control on production. Results of a preliminary economic evaluation showed that net revenues ranged from USD$ 2361.1-3488.9, the benefit-cost ratio from 1.47 to 1.62 and that the best and worse results were obtained by conducting two and one partial harvests. We conclude that the models are useful for analysing uncertainty of semi-intensive shrimp production incorporating partial harvesting.