Two methods of variance component estimation: Restricted Maximum Likelihood (REML) and Gibbs Sampling (GS), were used to analyze data consisting of 305d first lactation milk and fat yield and fat content of 47,574 cows calved from 1989 through 1996. Two three-trait linear models were applied, both including fixed effect of HYS, random effect of animal and error. One included fixed effect of age at calving class, and the other regression on age at calving. The estimates of genetic correlations and heritabilities were higher when Gibbs Sampling was used; the phenotypic correlations did not depend on the method applied (GS or REML). The heritability estimates for milk yield were 0.28-0.29 by GS and 0.24 by REML. For fat yield they were 0.24 with GS and 0.19 with REML. For fat content they were 0.44-0.45 and 0.37, respectively. The biggest difference between h² estimates was found for the fat content. The magnitude of heritabilities was not influenced by the linear model used (with age classes or regression on age).