Genetic evaluations of Holstein bulls from the US were matched with Canadian, Italian, Mexican, and Netherlands evaluations for the same bulls. Conversion equations for milk yield were computed by least squares, Goddard, and Wilmink methods. Accuracy was assessed by splitting data and applying equations developed from one subset to the other subset. Methods were judged by mean differences between actual and converted evaluations and standard deviation of that difference. Imperfection of conversions appeared to be due to inherent characteristics (variation and bias) of data rather than to inadequacy of conversion methodology. Least squares was slightly better than other methods but is not recommended by the International Bull Evaluation Service. The Goddard method was generally superior to the Wilmink method, but data often are not available for its application. A variation of the Goddard method was equal in accuracy to the Wilmink method. Daughter yield deviation as both dependent and independent variables was examined for only one data set and was little different from the Goddard method. Indirect equations were quite accurate for US to Mexico and US to the Netherlands but much less accurate for US to Italy conversion. Indirect conversions still would be useful until evaluations of bulls in common allow for direct conversions. For all three countries, a variation on indirect methodology was slightly superior to the usual indirect equations.