Summary Milk composition is affected by many environmental and inherited factors. To date there are no feeds, additives, or feeding practices that will practically and profitably change milk composition of dairy cows. The increased level of feeding of high-producing cows under some conditions of relatively high milk prices and low grain prices may come close to this objective, however. Overfeeding of protein, from accepted standards, will not increase SNF or milk protein content, but underfeeding protein will reduce milk yield. Pelleted, finely ground, or very high concentrate rations may drastically reduce milk fat content, often without a compensatory increase in other components or yield. Characteristic changes in rumen fatty acid concentrations are associated with these milk composition changes. Underfeeding total nutrients will reduce milk yield, SNF, and protein content. Recent and continuing research in rumen metabolism and the pathways whereby ration and environment influence milk composition is of great importance to future understanding of both genetic and management aspects of the problem. Season of the year, age of the cow, stage of lactation, and gestation all have an important effect on milk composition. Breeds differ in average milk composition and between individual cows within the breed there is even greater variation. Groups of paternal half-sisters differ significantly in mean values of milk fat, SNF, and protein content. The heritability estimates currently available indicate selection should be effective in changing milk composition. Further, the correlations between fat, SNF, or protein contents are not so high as to indicate their relationship could not be changed within reasonable limits over a long period of time. We need improved estimates of genetic and environmental parameters; these should be forthcoming from the regional milk composition projects. The problem of defining, in terms of nutritional and economic considerations, the optimum milk composition for the most efficient production of food needs serious study. Practices that increase total yield will usually decrease the per cent content of individual components slightly, but will result in the greatest net yield of each component. The expansion of SNF or protein-testing programs to gain research data and to prove sires for artificial insemination use seems very important. When definite goals become more clearly established, certain breeds or herds may find it most desirable to select for increased total yield. Others may give some attention to individual milk components and major attention to total yield, using an appropriate selection index to combine the objectives most efficiently.