Milk producers in the UK are paid according to their production of milk fat, protein and lactose and are also subject to a production quota which includes an element restricting fat production. The ability to predict the production of milk solids consequent on the feed inputs and animal state is therefore important. Results reported previously (Rook, Sutton and France, 1990) showed that multiple regressions of milk constituent yields on feed variables and animal characteristics such as parity and postcalving liveweight fitted poorly, with R2 ranging from 0.35 for lactose yield to 0.52 for protein yield. In practice, however, previous yields are often available and can be used to improve the fit of these empirical models. This study investigated the use of models including previous yields.