For four sub-populations of Atlantic salmon ( Salmo salar) in Norway, additive, dominance and additive by additive genetic variances and inbreeding depression were estimated for body weight at slaughter after two years in sea-water. The data used were from a hierarchical mating design, in which each sire was mated with two to seven dams. In each generation, 104 to 206 full-sib groups and 18 to 110 paternal half-sib groups were produced. Data from each sub-population covered three successive generations with pedigrees traced one generation further back to the base population. In the four sub-populations, the number of fish recorded ranged from 51 969 to 62 162 offsprings of 96 to 187 sires and 386 to 525 dams. A sire–dam model was applied separately to each of the sub-populations. In addition to the random additive, dominance and additive by additive genetic effects, the model also included the random environmental effect common to members of a full-sib group, a covariate of body weight on inbreeding coefficient, and a fixed subclass effect of year by farm by cage by sex. Variance components were estimated using an algorithm based on the tilde-hat approximation to REML with solutions for random effects obtained by iteration on data. In each of the four sub-populations, dominance and additive by additive genetic variances were at least equal to the additive genetic variance in magnitude. In every case, the inclusion of nonadditive genetic effects in the model lead to a 50 to 79% decline in the estimates of additive genetic variances, while reductions in the error variances were negligible. Estimates of common environmental effects ranged from 4 to 6% of the total phenotypic variance. Estimated inbreeding depression in the analysed data sets ranged from 0.6 to 2.6% per 10% increase in inbreeding coefficient.