Scientific surveys are widely used for stock assessment, but the estimated population parameters are based on the size-at-age relationship and age structure derived from a small subsample of the catch that is aged. This calls for an assessment of subsampling strategies, especially when population’s life history traits are spatially structured. In the Eastern Bering Sea, Pacific cod (Gadus macrocephalus) size and age are spatially structured, with younger and smaller individuals being more abundant at shallower depths. We conducted parallel subsamplings during Pacific cod surveys to compare two contrasting subsampling strategies: length-stratified and random. Geographical heterogeneity of Pacific cod length resulted in divergent estimates of ages between subsampling strategies. When this spatial variability was taken into account to estimate population parameters, random strategy provided more accurate mean and modal size-at-age and estimated age structure. Bias in the length-stratified subsampling arises from the poor efficacy in capturing the geographical patterns of size observed in the population. However, combining age data samples from multiple years helps to minimize the divergences between the two strategies.