A total of 296,627 lactation records which include milk, fat, protein yields (leche120, KgG120, KgP120) and fat and protein content (%G120, %P120) belonging to 102 herds of Manchega sheep breed were used to estimate genetic parameters for milk traits. Because the higher number of data available, random samplings from the initial population were carried out. A multivariate animal model with repeated measures was used to estimate variance conponents. Model included comparison group herd-yearseason of lambing, lambing number-age at lambing, number of lambs born and the period of time between lambing and the first test-day, as fixed effects and the additive genetic and the permanent environmental as random effects. Estimated heritabilities were 0.16, 0.13, 0.14, 0.09 and 0.28 and repeatabilities were 0.37, 0.31, 0.35, 0.15 y 0.41 for leche120, KgG120, KgP120, %G120 y %P120 respectively, which were lower than those found in other dairy sheep breeds and in previous studies in Manchega breed. Results from samples were quite similar among them and with the results obtained from univariate animal models for each single trait using all data available and no lost in precision using random samplings with a multivariate model were reported. Genetic correlations were high and positive between yields and moderately positive between %G120 and %P120, and negative or null between yields and composition. Because the economic importance of %G120 and %P120, it would be interesting to include both traits in the breeding objective.