Generalization of the polymorphism information content (PIC) index to represent marker informativeness (MI) for a three‐generation F~2~ design requires that two additional sources of non‐informativeness be added to the PIC formula: the probability of matings between like‐heterozygous F~1~ individuals, of which one is non‐informative; and that of matings between like‐heterozygous F~1~ individuals, which are both fully informative but where line of origin of the same alleles is reciprocal. Given the dense marker‐maps currently available for some species, this F~2~ informativeness parameter constitutes the natural criterion for marker selection in F~2~ designs, and two computer programs to predict MI from grandparental marker‐genotypes were developed for an F~2~ population originating from two divergent selection lines of outbred mice (F ∼ 0.2). A total of 403 markers had been genotyped for the F~0~ grandparents (n=31), and 14 markers had also been genotyped in the complete pedigree including 559 F~2~ individuals. One program was based on assumptions of random‐mating (RM), while the other (PED) accounted for the pedigreed mating structure. For the 403 markers, the correlation between MI from RM and from PED was 0.95, and the average deviation between the two predictions was 0.005 MI units (MI ranged from 0 to 1). Correlations between predicted and realized MI for the 14 fully genotyped markers were 0.97 for PED and 0.94 for RM, while the corresponding average of deviations between predicted and actual values were 0.01 and 0.04, respectively. Absolute deviations from realized MI never exceeded 0.09 and 0.16 for PED and RM, respectively. Simulated optimization of the mating system to maximize average MI of 28 markers on one chromosome led to improvements in the range of 15–20% average MI (0.07–0.09 MI units). The degree of relative advantage conferred by the F~2~ generalization of the PIC index over the traditional index was found to be of minor significance.