1. Using data from 396 breeding attempts over an 8-year period, we investigated age- and stage-specific survival rates and their modifying factors in a closed island population of the New Zealand stitchbird (or hihi, Notiomystis cincta Du Bus). 2. Survival probability generally increased over time; however, at each life-history transition, survival in the new stage started lower than at the end of the previous stage, creating a 'saw-tooth' function of age-related survival. 3. The probability of an egg hatching was low (0.73 +/- 0.01): most likely a consequence of genetic bottlenecks previously endured by this population. There was strong support for a positive relationship between hatching rate and the subsequent survival of the female parent, and hatching success declining for females > 4 years old. 4. Nestling survival probability increased as a function of brood size and days since hatching, and decreased relative to daily maximum ambient temperature and hatching date. Support for models including ambient temperature was greater than for other covariates, with the majority of this temperature-mediated survival effect being restricted to the early nestling stage. 5. Fledglings had low survival rates in the first two weeks after leaving the nest, with post-fledging survival related to the fledgling's mass. Two months after fledging, juvenile survival probability plateaued and remained relatively constant for the following autumn, winter and spring/summer breeding season. There was no effect of sex or season on adult survival probability. However, there was strong support for age-specific variation in adult survival, with survival likelihood increasing during the first four years before showing evidence of a senescence decline. 6. Within-stage survival increases were likely related to stage-specific selection pressures initially weeding out individuals of poorer phenotypes for the environment specific to each life-history stage. Such a mechanism explains the initial high mortality at life-history transitions; a well-adapted phenotype for one stage may not necessarily be so well adapted for subsequent stages. These patterns are not only valuable for examining life-history theory, but also for understanding the regulation of vital rates in an endangered species and providing a basis from which better population management models and harvesting regimes can be derived.