An integrated stock assessment for the Patagonian toothfish (Dissostichus eleginoides) for the Heard and McDonald Islands (CCAMLR Division 58.5.2), using CASAL and data consisting of multi-year random stratified trawl survey (RSTS) abundance estimates by length bin, commercial catch-at-length data, standardised CPUE series for the trawl grounds, and tag releases and recoveries by length bin, is described. The annual surveys are spatially representative of the main plateau, where juvenile fish are found, but are of relatively low intensity in effort compared to the commercial shots. In contrast, the commercial shots are very restricted in space, consisting of three main grounds. The model implemented in CASAL is a simple, single-sex, single-area population model, but spatial complexity in the fishery was modelled using separate fishing selectivity functions for each ground by gear (trawl and longline) combination. Various combinations of dataset weighting were investigated using haul-level estimated effective sample sizes with, additionally, iteratively estimated process error for the catch-at-length data and the inclusion versus exclusion of the tag data. A key uncertainty is the number of ages fully selected by the main survey series. With all the data included in the model, age-4 and 5 fish are fully selected, whereas when the survey data have greater influence and without the tag data, the selectivity of these ages was reduced. The method of quantifying process error used in this assessment removes ‘systematic lack-of-fit’ (SLOF) from population/fishery model predictions. Extension of this method of estimating process error to RSTS abundance data and commercial catch CPUE data is given, but incorporation of process error for the RSTS data was not considered appropriate, since SLOF could not be removed to an acceptable degree. The issues concerning the effect of the tension between survey data and mark–recapture data on parameter estimation are discussed.