Fishery managers need robust ways of objectively estimating the quantitative composition of fish stocks, by species and age-class, from representative samples of populations. Dual-frequency identification sonar data were used to first visually identify fish to a broad taxon (Salmonidae). Subsequently, kernel-density estimations, based on calibrated size-at-age data for the possible component species, were used to assign sonar observations both to species (Atlantic Salmon Salmo salar or Brown Trout Salmo trutta) and age-classes within species. The calculations are illustrated for alternative sets of calibration data. To obtain close and relevant fits, the approach fundamentally relies on having accurate and fully representative subcomponent distributions. Firmer inferences can be made if the component data sets correspond closely to the target information in both time and space. Given carefully chosen suites of component data, robust population composition estimates with narrow confidence intervals were obtained. General principles are stated, which indicate when such methods might work well or poorly.