AbstractVariance components for litter size (total number of piglets born) were estimated from Danish purebred Landrace and Yorkshire litters by restricting maximum likelihood. The data were collected from the national Danish breeding programme and consisted of 19 666 litters in Danish Landrace and 29 336 litters in Danish Yorkshire. Four different analyses for litter size were conducted within breed. In the first two, genetic groups were included in the model in order to account for the importation of animals from other countries; in the other two, genetic groups were removed from the model. Within each case, herd-year-type of insemination effects were fitted as fixed (H-fixed models), or herd-year-season-type of insemination effects were fitted as random (H-random models). Estimates of heritability ranged from about 0·11 to 0·14 in Landrace and from 0·10 to 0·11 in Yorkshire. Variance due to herd-year-season-type of insemination ranged from 0·029 to 0·041 of total variance, values somewhat lower than those obtained for non-genetic permanent effects. In order to compare the four models, data were divided into different subsets, and records from one subset were predicted using parameters estimated from the other subset. Both the correlation between observed and predicted values, and the mean squared error of prediction indicated that predictive ability was higher in the case of H-random models. There was no evidence that genetic groups improved the predictive ability for litter size. However, group effects affected inferences about genetic trend, particularly in Landrace, where genetic group composition changed consistently over the years.