Traditionally, agricultural forecasts, whether for the coming year or several years into the future, have been based on assumptions of normal weather and trend crop yields. That weather is seldom normal and that yields seldom fit trends are well recognized. However, relatively little attention has been given to projecting crop yields stochastically even though computer capacity and software programs are available to do so. One reason is that the task is more challenging than to assign standard deviations to various crop yields and simulate normal distributions using random number generators. For one, deviations of crop yields from trends may be correlated especially if the locations of the crops overlap such as is the case with US corn and soybeans. To model US agriculture, those correlations must be taken into account. Secondly, deviations of crop yields from trends may not be normal. Typically, crop yield deviations are skewed to the low side, with yields lower in poor crop years than higher in favorable crop years. This paper demonstrates how computer software programs can be used to generate probability distributions of yields taking into consideration correlations among crops and non-normality in distributions. Included are thirteen crops and crop aggregates with global coverage of coarse grains, wheat and oilseeds. Probability forecasts are made for 2006 and illustrated for US corn, soybeans and wheat.