We consider estimation of the magnitude of incidental fisheries 'bycatch' for two petrel species, sooty shearwaters (Puffinus griseus) and short-tailed shearwaters (Puffinus tenuirostris). There are clear statistical advantages in estimating bycatch for abundant species such as these, and our results may also guide the conservation and management of rarer species. We used fisheries statistics and observer data to estimate retrospectively the total numbers of sooty and short-tailed shearwaters bycatch in seven large-scale pelagic North Pacific driftnet fisheries between 1952 and 2001. Sensitivity analysis greatly simplified estimation of uncertainty by identifying four driftnet fisheries to be of particular importance in determining the magnitude and precision of the estimated bycatch totals. We estimated that between 1.0 and 12.8 million (95% CI) sooty shearwaters were killed by driftnets between 1952 and 2001. For short-tailed shearwaters we estimated between 4.6 and 21.2 million (95% CI) over the same period. More precise estimation was hampered by the paucity of available observer data, lack of reported detail and inconsistencies among data sources. Estimates may be strongly biased because some dead birds are misidentified or drop out of nets before hauling, or because some records were of live captures that were subsequently released. Improved estimation of overall take and its impact on populations of seabirds requires standardisation of reporting, allowance for potential sampling bias, as well as a clearer definition of the sampling unit and underlying bycatch probability distribution model, and knowledge of potential compensatory changes in population parameters.