Multiple-peril crop insurance policies require statistical modeling of probability distributions of crop yields. Unfortunately, no single parametric distribution is likely to capture the true data generating process. Likewise, non-parametric approaches converge to the true distribution at a slow rate; yield histories are often of modest size. Recognizing these shortcomings, model and forecast combination are now being applied in crop insurance settings. Model and forecast combination avoid the dangers inherent in selecting a single model for the predictive yield distribution. The component models for the combination can be selected ad-hoc or based on the idea of distributional similarity. Crop yields are spatially correlated, so the model for one insured unit may be related to another, and can then be used in the pool of potential models for the combination. We briefly review the literature on model and forecast combination and its application in agricultural insurance settings. We then turn toward an empirical application involving crop yield insurance for major row crops in the U.S. Southeast. A variety of individual models and combinations are estimated at the county level. Insurance premiums and premium rates are calculated from the estimated distributions. Implications of model and forecast combination for insurance rates, premiums, and government subsidies are discussed. We conclude by suggesting future research in this area.