This study investigates the nature of the revision process of USDA corn and soybean production forecasts over the 1970/71 through 2002/03 marketing years. Nordhaus' framework for testing the efficiency of fixed-event forecasts is used. In this framework, efficiency is based on independence of forecast revisions. Both parametric and non-parametric tests reject independence of consecutive forecast revisions. Positive correlation and consistency of directional changes in forecast revisions suggest that these forecasts are "smoothed." Estimates of the impact of smoothing on forecast accuracy show that correction for smoothing may result in economically meaningful improvements in accuracy.