Comparison of computation time between single-trait and multiple-trait evaluations showed that with the use of the canonicat transformation associated with multiple diagonalization of (co)variance matrices, multiple-trait analysis for milk, fat and protein yields is not more expensive than three single-trait analyzes. Rank correlations between breeding values for 54,820 cows with records (for their 1,406 sires) estimated with the single-trait and multiple-trait models were over .98 (.99) in fat yield and over .99 (.99) in milk and protein yields. The relative gain expressed as reduction in mean prediction error variance was 3% (1%) in milk yield, 6% (3%) in fat yield, and .4% (.2%) in protein yield for cows (for sires). Relative genetic gains were 3% (1%), 6% (2%) and .5% (.2%) respectively in milk, fat and protein yields for cows (for sires). The use of multiple-trait models bas therefore the advantages of improved precision and reduced selection bics. Multiple-trait analysis could be extended for the analyzes of test-day records. Results show that this or similar multiple-trait animal model could be implemented immediately in Belgium at low computing cost, using the proposed algorithme and could be the first step to new, more advanced evaluation methods.