African Crop Science Journal 2014, Vol. 22, Issue Supplement s4: pp. 941 - 950Seasonal rainfall variability, particularly the uncertainty with respect to the direction and extent that variability will assume in a given season, forms the greatest source of risk to crop production in semi-arid areas of Ethiopia. Equipping vulnerable communities, in advance, with the expected date of onset of a cropping season, is crucial for smallholder farmers to better prepare to respond and manage the uncertainties. Therefore, rainfall prediction, particularly development of models that can foretell the date of onset of next cropping season is crucial in facilitating strategic agronomic planning and tactical management of in-season risks. A twenty-four-year climatic data study was conducted for Melkassa Agricultural Research Centre (MARC) in semi arid Ethiopia, to develop onset date prediction models that can improve strategic and tactical response farming (RF). A sequential simulation model for a build up of 15 to 25 mm soil water by April 1st, was conducted. Simulation results revealed a build up of soil water up to 25 mm, to be the most risk-wise acceptable time of season onset for planting of a 150-day maize crop. In the context of response farming, this was desirable as it offers the opportunity for farmers to consider flexible combination production of maize (Zea mays L.) varieties of 120 and 90 days in the event of failure of earliest sown 150-day maize crop. Thus, to allow for flexible combination production of the three maize varieties, predictive capacity was found crucial for April onset of the next crop season. Accordingly, based on the consideration of pre-onset rainfall parameters, the first effective rainfall date varied considerably with the date of onset of rainfall. Regression analyses revealed the first effective rainfall date to be the best predictor of the date of onset (R2 = 62.5%), and a good indicator of the duration of next season (R2 = 42.4%). The identified strategic predictor, the first effective rainfall date, enabled prediction of time of season onset and season length by a lead time of two to three months. This markedly improved Stewart’s RF. The date of onset of the next crop season was also found to be a useful predictor of season duration (R2 = 87.3%). Strategic agronomic planning should be adjusted according to the first effective rain date, and tactically according to what date of rainfall onset informs us about expectations in the duration and total season water supply.