Six popular crop yield distributions are compared to determine which best describes yield fluctuations out-of-sample. For 183 crop and county combinations, each distribution is estimated and ranked according to its log-likelihood function observed at out-of-sample observations. A semiparametric model dominates the contest for most crops and counties, likely due to its flexibility and treatment of heteroskedasticity. Most other models ranked lower because their variance equation performed poorly out-of-sample.