Microsatellites allow pedigrees to be redrawn in groups of mixed families. However, genotyping costs are still high and experiments deserve new optimizations based on total number of genotyped offspring. Using stochastic simulations, this paper compares and optimizes three factorial or partly factorial mating designs for inference on heritability: FF: full factorial: s sires, each mated with s dams (s2 families) FD: s sires, each mated with two dams (2 s families). BH: described by Berg and Henryon wA comparison of mating designs for inference on genetic parameters in fish. Proceedings of the Sixth World Congress on Genetic Applied to Livestock Production, Armidale, Australia, 11–16th January, 27, 115.x: s sires mated with s dams: sire 1 with dams 1, 2; sire 2 with dams 2, 3 . . . sire S with dams 1, S (2 s families) A quantitative trait is simulated according to a strictly additive, polygenic model. Two levels for number of genotyped offspring (NOs300 or 1000) and three levels for true heritability (h2s0.1, 0.25 and 0.5) are considered. For each NO–h2 combination, all possible couples number of families (NF)/family size were studied to find the one giving the most precise estimation of heritability. Standard deviation of heritability was calculated over 5000 repetitions in each situation. In most cases, FF designs are less interesting than FD and BH ones. FD designs are more precise than BH designs, except for NOs1000 and h2s0.5. Optimum family size is similar for FD and BH designs and both NOs: 3–5 offspring per family for h2s0.5, 5–8 offspring per family for h2s0.25 and 12–20 offspring per family for h2s0.1.