Actual farm records were used to evaluate variation of crop yields for corn, wheat, and soybeans grown in Michigan. The assumption of yields being normally distribute with independent and constant variance was tested for all farms, farms characterized by soil groupings, and individual farms of a selected soil group. Sufficient evidence was not found to reject the assumption that yield variation is normally distributed with independent and constant variance when all farms were grouped together. Grouping farms by soil potential showed strong support for yields being independent and normally distributed. The assumption of constant variance had strong support for wheat, but was weak for corn. Analysis of individual farms for a selected soil group, suggested that farms having data sets with significant amounts of negative kurtosis or negative outliers could be better modeled by alternative distributions. Estimates for yield variation parameters were incorporated into planning tools used for predictive analysis.