Establishing whether pre-industrial societies caused significant harvesting impacts on fish stocks is often hindered by the paucity of historic evidence. Some archaeological assemblages contain information on the sizes and/or species of individuals in the catch, but this does not provide any direct evidence on the absolute size of the catch or comparative metrics. We develop a method for using size-frequency data to infer the intensity of fishing and the size-selectivity of the fishing gear in use. The model allows quantitative estimates to be made for the depletion of snapper populations relative to the unexploited pre-human biomass. We evaluate this method using six modern and five archaeological datasets from northern New Zealand for a key commercial and artisanal species, Australasian snapper or silver seabream (Pagrus auratus). Our method uses two models for the size selectivity of fishing: one S-shaped, representing mobile fishing gear such as trawls or seines, and one dome-shaped, representing static fishing gear, such as hooks, longlines, or gillnets. The results show that the estimated fishing intensity is lower, and the size of fish being caught is larger, in the archaeological datasets than in the modern datasets, as might be expected. Nevertheless, some of the archaeological datasets show evidence that is consistent with substantial resource depression and depletion of the largest fish in the population, while others suggest only light exploitation. The method allows the five archaeological cases to be rank ordered in terms of exploitation pressures and the relative orderings are further assessed using independent information on site chronology, stratigraphy, and recovery procedures (i.e., screen size). Other factors that can affect size-frequency data are briefly considered, but require additional environmental and taphonomic data that are not currently available. The results provided by our new method support the hypothesis that the depletion of large fish and capture of progressively smaller ones occurred in the pre-European era, albeit in spatially localized areas and at a much less severe level than in modern times. The model results also help identify potential biases in the archaeological assemblages and directions for further research.