Heritability coefficients as well as animals' breeding values (BVs) were estimated on an average information algorithm for raw milk yield (MY), log transformed (LMY), square root transformed (SMY), adjusted for mean (CM) and adjusted for standard deviation (CSD) heterogeneity of the herd-year-lactation (HYL) classes, MY. The data included 4100 lactation records from 2551 ewes of the Lesbos dairy sheep kept in 44 herds on the homonymous island. Employment of data transformation (log or square root) and adjustment for mean heterogeneity did not result in homogenous variances across the HYL classes. Only adjustment of records by regressing the standard deviation (S.D.) of the HYLi class on population mean S.D. resulted in homogenous variances. Heritability estimates (h 2 ) for MY, LMY and SMY were found to be of same magnitude (h 2 =0.19) while h 2 for CM was higher (h 2 =0.20). Adjustment of MY for heterogeneous S.D. decreased the heritability estimate (h 2 =0.16), as a result of proportionally higher reduction of the additive genetic than of the residual variance. Heritability coefficient has also been estimated for two production (low and high) and two S.D. (low and high) levels of the HYL classes. Heritability estimates between low and high production levels were found of comparable magnitude (0.18 vs. 0.22, respectively) while they were clearly discriminated between low and high S.D. level (0.09 vs. 0.29, respectively). Product-moment (Pearson) and rank (Spearman) correlations and proportion of animals selected before and after adjustment were also estimated between estimated BVs on MY, CM and CSD records. Correlation estimates were generally high (r=0.88-0.98) when all sires and all ewes of the population were considered. Heterogeneity of variance had, however, a remarkable impact on ranking and thus of animals selected in 5% top sires and dams. Correction for heterogeneity of S.D. caused a great proportion, i.e. 25% and 32% of top sires and dams, respectively, to be excluded from selection when compared to the homogenous variance scenario. However, selection on a wider basis, e.g. sires with positive BVs and 20% of best dams, resulted in lower proportions of animals excluded. This was also the case for the effect of mean heterogeneity on the proportions of top sires and ewes selected. Investigation of the sources of heterogeneity of the HYL S.D.s showed that factors such as herd, lactation number, number of observations and production level are associated with heterogeneity of variance. Production year as well as genetic level did not significantly affect heterogeneity of variance (P>0.05). These above factors explained about 53% of the variation of S.D.s of the HYL classes.