The EPIC model (Williams et al, 1989) is commonly used to simulate the wheat crop in a crop rotation, thoughit overestimates grain yield in dry conditions. A new version of the model (EPICPHASE) has been developed by INRA in Toulouse to improve the model performance in drought-prone environments. The refinements introduced were: 1) a better description of the soil water extraction pattern; 2) a response of harvest index to water and nitrogen stress as a function of phenological phase. This study describes this new version and proposes the corresponding calibration for winter soft wheat. Literature and various experiments (lysimeters, pluri-annual field testing) were used to determine the optimal set of parameters. Functions describing wheat growth without limiting factors under water constraint were calibrated (ie, biomass accumulation and partitioning, leaf area and root depth, water uptake). The model was tested in Auzeville (Haute-Garonne, SW France) during two successive dry years (1989 and 1990) differing in their drought pattern, and on a range of fields differing in water and nitrogen supply. Referring to the extreme variability of the test situations, the predictions of the model were satisfactory for water budget, biomass, yield and nitrogen uptake, with mean relative errors of 8-18%, depending on variable and year. The sowing-heading duration was predicted with a root mean squared error of 59 degree.days. The EPICPHASE-Wheat model should be quite valuable for assessing the impact of different management schemes in rainfed or irrigated wheat crops and for defining optimum strategies.