In October of 2002, several countries in Southern Africa faced dramatic food insecurity crises, with more than 10.8 million people requiring food aid. Compounding concerns were forecasts of an El Niņo event in late 2002 and early 2003, a global climate phenomenon historically linked to drought in Southern Africa. Climate analyses by FEWS NET scientists also showed that above-average rains in October typically presage below-normal precipitation during the ensuing January-February-March period, the heart of the growing season. El Niņo conditions intensified, and October rains in the region were well above normal. This experience of above-normal October rains and the onset of El Niņo in the midst of a food security crisis consequently prompted the development of a new seasonal forecasting approach. The method, matched filter regression, used regional October atmospheric circulation data (upper and mid-level winds and precipitable water) in conjunction with October sea surface temperatures from the Indian, Atlantic and Pacific Oceans to forecast rainfed agricultural outcomes for the harvest five months later. The technique was used in November 2002 to forecast April 2003 (end-of-season) cumulative crop water satisfaction for maize. We describe the forecast technique, cross-validated results based on its application over the period 1972-2003, and the FEWS NET forecast for the 2002-2003 season. A post-harvest evaluation based on observed data through end-of-season showed that the forecasting technique worked well, correctly identifying outcomes for 8 of 12 crop areas modeled, including several severely drought-stricken regions. As a consequence of these climate analyses, food security analysts were given an early opportunity to factor the high likelihood of a second consecutive year of drought into their planning for humanitarian assistance activities in 2003.