A multi-variate animal model Best Linear Unbiased Prediction (BLUP) on milk, fat and protein yields in different lactations, as different traits, is the optimum method for genetic evaluation in dairy cattle for production. However, this is computationally demanding and usually a repeatability model is performed separately for milk, fat and protein yields. A simulation study by Visscher (1991) showed that the repeatability model on canonical transformed yield rather than on observed yield was a better approximation to a multi-variate analysis using selection index. The aim of this study is to verify, using an animal model, the efficiency of the repeatability model relative to a multi-variate analysis on observed or transformed yields.