Population modelling software was developed that captures the dynamics of spatial heterogeneity of the population along with age structure, movement and reproductive stage transition in a holistic framework. Using this software, three spatially explicit age-structured models were developed for the Antarctic toothfish (Dissostichus mawsoni) population in the Ross Sea region and used as operating models to evaluate the performance of single-area assessments. The models were generalised Bayesian population models and were optimised by fitting to fishery observations. Movement was parameterised using preference functions based on spatially discrete environmental layers. The shapes of the preference functions were established through iterative model testing whilst the parameters defining the preference functions were estimated within each model. The spatial structure of the models divided the Ross Sea region into 189 equal-area (24 000 km²) cells. The underlying spatial distribution of the population was either restricted to the 65 cells historically fished, to the 120 cells containing habitable depths, or to the entire Ross Sea region (all 189 cells). Estimates of movement rates were consistent with the results of tagging studies and fits to the other observations (age, reproductive development) were adequate. These operating models were then used to investigate potential biases of the current single-area stock assessment. Simulations based on the three spatial distribution scenarios suggested that the current single-area stock assessment is biased low by 19–43%.