The traditional approach to projecting the distribution of farms by size uses a Markov model with stationary (constant) transition probabilities. While a useful tool for extrapolation of current trends, the stationary Markov approach cannot model the impacts on farm structure of varying economic and social causal forces. Data are now available for developing Markov models with nonstationary transition probabilities. A simple nonstationary Markov model of U.S. farm structure is described and estimated, and its performance in predicting actual changes in farm numbers and sizes through 1986 is assessed. Further issues in the development of conditional projections of farm structure are discussed.