Obtaining reliable population estimates is crucial to monitoring endangered species and developing recovery strategies. The palila (Loxioides bailleui) is an endangered seed-eating Hawaiian honeycreeper restricted to the subalpine forests of Mauna Kea, a volcano on the island of Hawai‘i, USA. The species is vulnerable to extinction primarily because >90% of the population is concentrated in <30 km² of habitat on the western slope of this high, dormant volcano. Annual surveys of the palila population have been conducted for ecological, legal, and other purposes since 1980. Because refinements to sampling protocols and analytical methods have evolved, we examined means of adapting the monitoring program to produce comparable estimates of abundance over the past 25-year period and into the future. We conducted variable circular plot surveys during the nonbreeding season (Jan–Mar) and this used data to obtain estimates of effective detection radius and annual density with Distance 4.0, Release 2. For comparability over the time-series, we excluded from analysis the data from new transects. We partitioned the 25-year data set (1980–1996 and 1997–2004) into 2 separate analyses because, beginning in 1997, observers received more training to reduce their tendency to estimate distances to 5-m intervals. We used geographic strata in the analysis of recent surveys because changes in habitat may have invalidated the density-based strata used previously. By adding observer and year and observer and time of day as co-variables, we improved the model fit to the 2 data sets, respectively. Annual estimates were confounded by changes in sampling methodology and analytical procedures over time. However, the addition of new transects, increased training for observers, and use of exact distance estimates instead of rounding also improved model fit. Habitat characteristics and behavior of palila that potentially influenced detection probability, sampling, analysis, and interpretation were regeneration of trees in response to reduced numbers of introduced browsing mammals, seasonally variable rates of vocalization, non-territoriality, and resource-tracking along an elevation gradient. We believe our adaptive approach to analysis and interpretation of 25 years of annual variable circular plot data could help guide similar long-term monitoring efforts.