Family mapping is based on multiple segregating families and is becoming increasingly popular because of its advantages over population mapping. Athough much progress has been made recently, the optimum design and allocation of resources for family mapping remains unclear. Here, we addressed these issues using a simulation study, resample model averaging and cross-validation approaches. Our results show that in family mapping, the predictive power and the accuracy of quatitative trait loci (QTL) detection depend greatly on the population size and phenotyping intensity. With small population sizes or few test environments, QTL results become unreliable and are hampered by a large bias in the estimation of the proportion of genotypic variance explained by the detected QTL. In addition, we observed that even though good results can be achieved with low marker densities, no plateau is reached with our full marker complement. This suggests that higher quality results could be achieved with greater marker densities or sequence data, which will be available in the near future for many species.