Monthly rainfall was correlated with shire wheat yields across the Australian wheat belt and an average correlation coefficient determined for each month. Based on this varying usefulness of rainfall throughout the year, a rainfall index that weights district rainfall over the wheat belt was found strongly to relate to state and national wheat yields. Accuracy of the model was improved by filtering out insignificant and excess rainfall according to the broad scale water holding capacities of regional soils. As one moves clockwise around Australia from the east the soil is less able to store water and truncated winter rainfall is a major problem on poorer soils. Assuming actual rainfall up to the forecast date and average rainfall beyond, hindcasted yields in 1988 and 1989 were within 10% of the final figure 3 months in advance of harvest. A marked trend to early sowing and higher yielding varieties, coupled with increased inputs caused the model to underestimate in 1990, 1991 and 1992. However, at the end of the year model predictions equalled, or were more accurate than official predictions in 4 out of 5 years. Large-area Australian yields appear to be a function of both rainfall amount and distribution, as well as the time of sowing and extent of frost in critical periods.