Forecasting banana supplies is an important priority both for growers and for the local marketing organisation. Achieving this requirement is complicated by the floral ontogeny of banana which induces flowering and fruit production throughout the year. At the two experimental localities Burgershall and Levubu, in the Eastern Transvaal, totals of 2571 and 2057 ratoon ‘Williams’ banana bunches, respectively, from uniform plantations, were analysed over a 3-year period. Mean flower emergence to harvest interval and mean bunch mass were calculated for 24 flowering periods of 15 days each. Seasonal variations at both localities were substantial and were related to temperatures prevailing at flower initiation and during fruit development. At Burgershall, the shortest flowering to harvest interval was 110 days (early summer flowering) and the longest 204 days (autumn flowering). Bunch mass varied from 57.8 kg (July flowering) to 37.2 kg (October flowering), the latter bunches resulting from winter flower initiation. Levubu data differed substantially from Burgershall, although the seasonal trends were similar. A practical crop forecasting model is proposed which is based on new flower counts every 15 days, and the overall box/bunch potential of the grower. Using the experimental data base, the total boxes are forecast from seasonal yield adjustment factors, and harvest spread is forecast from the harvest distribution of experimental bunches for each flowering period. Each cultivar-locality combination must be forecast separately, and the model assumes year-to-year uniformity, farm-to-farm uniformity, optimum management, and an absence of major damage due to wind, hail or frost.