A method of obtaining estimates of covariance matrices from animal models in multitrait situations is put forward for further study. The method is based on approximating the prediction error covariance matrices of estimates of random effects in the model. The simplified approximations allow the estimation of covariance matrices from very large data sets, with different models for each trait and with missing observations on some traits. Comparison of the approximations with actual values in small data sets has not been done, and refinements to the approximations may be possible. Estimation will still require large amounts of computing due to the sheer number of animals that need to be included in such analyses

%V XIII. Plenary lectures, molecular genetics and mapping, selection, prediction and estimation. %P 441–444 %G eng