This paper describes the experimental and conceptual basis of a simulation model to predict the seasonal growth pattern of undefoliated Townsville stylo ( Stylosanthes humilis H.B.K.) on Tippera clay loam, an extensive soil type in the Tipperary land system, N.T., Australia. This tropical legume behaves as a self-seeding annual in the tropical savannah climate of this region. Experiments in two consecutive wet seasons with rainfalls of 705 mm and 913 mm provided the experimental data for the model, average rainfall for the region being 925 mm. Meteorological data required by the model are rainfall and pan evaporation. The available water content range of the soil is used to simulate water distribution through the profile. The ratio of rates of evapotranspiration to pan evaporation is related to dry matter of standing tops when water does not limit transpiration. The rate of evapotranspiration so calculated is then modified by the availability of water in the root zone (as simulated by the model) to give the actual evapotranspiration rate. Dry matter at any time is then inferred from actual evapotranspiration accumulated from the time of germination and establishment of the crop. The relationship between these two variables is dependent on phosphate nutrition. The model overestimates experimental yield, for example by 8% in a season with a rainfall well above average. It is desirable that model performance in seasons with well below average rainfall should receive further checking, but again indications of over-estimation were obtained. The model reflects current limitations in our understanding of phosphate availability and plant nutrition, and an indication of the possible importance of such limitations was given by discrepancies between prediction and the growth of two sequential sowings on land with no previous history of phosphate application.