Farm income is highly variable due to annual price and yield uncertainties. The federally subsidized crop insurance program is an important tool for managing this risk, and has grown from a relatively modest program to one that encompasses the majority of productive cropland in the country. The success of this program depends on identification of actuarially fair insurance premium rates, which in turn depends on accurate estimation of farm-level yield distributions. We use the confidential U.S. Department of Agriculture Risk Management Agency (RMA) panel dataset to estimate farm-specific distributions of yields and actually fair crop insurance premiums. Our ongoing work includes using the difference between our estimated actually fair premiums and RMA's to predict which insurance contracts farmers select. Ultimately, we will predict potential efficiency gains from using our empirical model for premium determination.