With the observer data from the National Marine Fisheries Service Pelagic Observer Programme (POP) and the logbook data from the US Atlantic pelagic longline fishery, we estimated the seabird bycatch in the fishery during 1992–2012. The POP observed 13 847 longline sets, with a total of 141 seabirds captured on 74 sets. The overall nominal catch rate was 0.0102 birds per set and 0.014 birds per 1000 hooks. We applied a random year effect model (RYEM) for analysis of the whole study region that includes 11 fishing zones. Extrapolating from the observed seabird bycatch, we estimated a total of 2255 seabirds captured on average (coefficient of variation CV = 14.72%) by the total fleet from 1992 to 2012. The highest estimate of seabird bycatch occurred in the middle Atlantic bight (MAB), followed by the northeast coast (NEC). Estimated seabird bycatch, by season, was higher in summer, fall, and winter than in spring. Longline sets targeting a mixed group of species caught the majority of the total seabird bycatch, and longline sets targeting swordfish and tuna also caught more seabirds than those sets targeting other species. To incorporate spatial variation into parameter estimation, allowing parameters to vary spatially, we applied a spatial expansion model (SEM) to data for the three fishing zones with the most observed seabird bycatch, the NEC, the MAB and the south Atlantic bight. When compared with the estimates from the RYEM (145–1049 seabirds with a CV of 16.4–23.5%), the SEM produced higher estimates (155–1489 seabirds) of the total seabird bycatch for each of these areas and a larger CV (19.1–65.4%). The RYEM may be appropriate for seabird bycatch assessment when spatial variation is not a concern; the SEM could be an alternative when observed data vary greatly over space.