This experiment was designed to quantify the relationships between feed chemistry and the proportions of rumen volatile fatty acids (VFA) across a wide range of feedstuffs. In an experiment using 11 rumencannulated sheep, 16 test feeds were fed at three different inclusion rates in rations that were based on grass silage. The 17 periods of the experiment were each 14 d long. Eight rumen samples were taken every 24 h on d 13 and 14 of each period from which the mean daily proportions of VFA were derived. The effects of an increase in the proportion of test feed in the total feed on proportions of VFA were significant. The observed proportions of VFA were related to the chemical composition of the total feed by principal component regression. The inclusion of the amount of feed offered and the ratio of test feed to total feed in these regressions did not improve their precision; these terms were not significant. The significant terms in the regressions were crude protein, starch, sugar, and cellulose (calculated by difference). The R2 values achieved for the regressions between acetate, propionate, and butyrate (molar proportions) and feed composition were 77.5, 68.0, and 87.3%, respectively. These regressions provided an apparently robust basis for predicting molar proportions of VFA from feed chemistry in feeds based on grass silage.