Abstract A practical model that considers both deficits and excesses as days of stress, AQUA, was developed and validated against crop yields from Northwestern Costa Rica. AQUA is based on daily water balance calculations, and inputs are: Daily precipitation ( P); averages of monthly potential evapotranspiration ( ETP); and crop and growth stage parameters that affect crop available water ( CAW) and crop ET, and determine the minimum amount of P that produces excess. Daily calculated monthly deficit (mm) and number of days in month with deficit were equivalent for different stations and CAW values. Simpler methods, P month/ ETP month and P month − ETP month yielded different relationships against days in month with either deficit or excess for the different CAW values considered. Deficit and stress (deficit + excess) indices, based on number of days, were validated against two planting dates of maize, and one each of rice and beans. Ten years of yield data were available for rice, only six for the other crops. Highly significant ( p < 0.01) correlations were found between the indices and yields of rice and second maize planting. Coefficients of determination ( r 2), the fraction (or percentage) of yield variations explained were, for rice, 0.83 for Deficit Index (IDef) and 0.90 for Stress Index (IStres), and for maize, 0.85 (IDef) and 0.87 (IStres). For first maize planting, r 2 values were 0.61 ( p < 0.1) and 0.67 ( p < 0.05) for IDef and IStres, respectively. Correlations against bean yields were poorer; nevertheless, r 2 values were 0.49 (N.S.) and 0.60 ( p < 0.1) for IDef and IStres, respectively. Though in a limited manner, considering excesses improved yield explanation in all cases, validating the methodology selected. The differences in yield dependance on water stress between crops, and factors that may induce underestimation of the effects of water excesses are discussed. The ability of AQUA to explain from 60 to 90% of yield variations even with limited yield data, the simplicity of the underlying assumptions, together with the ubiquity of the meteorological data required, should favor this kind of model over simpler ones. The use of a dynamic, microcomputer-based model like AQUA could represent a powerful and low-cost tool for regionalization, yield prediction, and rainfed crop management in the tropics.