We used quantitative fatty acid signature analysis (QFASA) to predict the long-term diet of New Zealand (NZ) sea lions (Phocarctos hookeri) incidentally caught in the NZ arrow squid (Nototodarus spp.) fishery. The QFASA model used fatty acid (FA) profiles based on 82 blubber samples of NZ sea lions bycaught between 2000 and 2006. First, the model was optimized by a series of simulations for which 1 model parameter—6 different sets of calibration coefficients (CCs) from different pinniped species and feeding regime, 2 sets of FAs, and the consideration of individual prey values, or mean prey values—varied each time. The best-fit parameters were those giving the lowest Kullback–Liebler distance values. Second, these parameters were used in a model to estimate the diet of NZ sea lions. QFASA was highly sensitive to the set of CCs applied. Across years the most important prey estimated with the best-fit CCs were southern arrow squid (Nototodarus sloani, 18–28% mass), hoki (Macruronus novaezelandiae, 10–27% mass), rattails (Macrouridae, 7–27% mass), and possibly scampi (Metanephrops challengeri, 1–19% mass). Despite the uncertainty on the accuracy of the match between the best-fit CCs used and the true FA metabolism of NZ sea lions, the variation of prey estimated among years was highly consistent with the trends of commercial catches during the same period, providing some confidence in the present QFASA predictions. The most important estimated prey were demersal species living mainly at depths >200 m that NZ sea lions encounter on the slopes of the Auckland Islands shelf. Our study emphasized the importance of these areas for bycaught NZ sea lions over the 1st half of the lactation period.