The problem of analyzing a sequence of daily irrigation decisions utilizing weather forecast information is formulated for the case of lettuce grown in central New York state, and solved using a stochastic dynamic programming algorithm. The crop response is represented using a simple but physiologically-based model of lettuce growth [van Henten, E.J., 1994. Validation of a dynamic lettuce growth model for greenhouse climate control, Agric. Sys. 45, pp. 55-72], modified to allow the stomatal conductance for CO 2 to depend on a simple soil moisture budget. A negative crop response to prolonged wet soil conditions combined with warm temperatures is also included in the crop model. Operationally available precipitation and temperature forecasts are incorporated in a way that preserves the effect of time correlation in the weather. The results suggest that irrigation is quite viable even in the relatively humid climate of New York, with the economic value of irrigation (scheduled according to a conventional, non-optimal rule) vs. no irrigation estimated at approximately US$4000US$5000 per hectare per year for lettuce. Optimal use of weather forecasts to schedule irrigations is estimated to provide additional value of approximately US$1000 per hectare per year, much of which is derived from avoiding crop damage due to excessive soil moisture. © 1998 Elsevier Science B.V.