The use of marker-based pedigrees is increasing in aquaculture breeding, and obtaining high assignment rates is necessary for practical use of this methodology. In this paper, we used 12 real parentage assignment datasets from three species (European sea bass Dicentrarchus labrax, common carp Cyprinus carpio and rainbow trout Oncorhynchus mykiss) to investigate the relationships between theoretical, simulated and real assignment power. We found out that there was a large decrease between the theoretical and the observed values, which we modeled in four independent steps: 1) from theoretical values to population-wise simulations (−2.8% on average), 2) from population-wise simulations to true parent set specific simulations (−2.6% on average), 3) from true parent-set specific simulations to observed values in offspring with valid genotypes at all loci (−0.5% on average) and 4) from observed values in offspring with valid genotypes at all loci to observed values in all offspring sampled (−2.4% on average). For all steps, we provide a regression equation which models the loss of assignment power, or at least a maximal practical value for the loss of assignment power. Finally, equations are provided to model the expected true assignment rate from the theoretical assignment power or from the combined exclusion probability of the loci used. They show that the expected true assignment rates are considerably lower than the theoretical ones, for example achieving 99% true assignment requires a theoretical assignment power of 99.999996%, while 95% already requires 99.9989% theoretical assignment rate.