Technical Abstract: A method of accounting for differences in variation in components of test-day milk production records was developed. This method could improve the accuracy of genetic evaluations. A random regression model is used to analyze the data, then a transformation is applied to the random regression coefficients. Comparing evaluations with and without this transformation, rankings of animals were similar but some re-rankings occurred. A correlation analysis showed a reduction in the effect of production level on the variation in regression coefficients indicating the success of the transformation. The method suggests a number of innovative solutions to problems related to heterogeneous variance structures.