Data on 4477 mares born between 1932 and 1995 from six studs were recorded. Three exterior traits were included: height at wither, chest girth, circumference of cannon. (Co)variance components were estimated by the DFREML procedure on the basis of four linear models. Model I included the fixed effects: period of birth, place of birth, breed effect (as covariable) and random effects: direct additive genetic and residuals. Model II was expanded by taking into account the maternal additive genetic effect. Third model included cytoplasmic effect as fixed (model IIIa) and random (model IIIb). Direct heritability estimates varied from 0.4 to 0.6 whereas maternal heritability estimates ranged between 0.01 and 0.25, cytoplasmic heritabilities were close to zero. Negative correlations between direct and maternal effects were estimated for all traits under study. Akaike Information Criterion suggested the highest adequacy of model I. However, other criteria (likelihood ratio test and error variance estimate) and negative covariance between direct and maternal genetic effects indicate that maternal effects should be included in the genetic evaluation of horses. In general, genetic trends were positive for the traits studied.