Questions: Do long-term observations in permanent plots confirm the conceptual model of Metrosideros polymorpha cohort dynamics as postulated in 1987? Do regeneration patterns occur independently of substrate age, i.e. of direct volcanic disturbance impact? Location: The windward mountain slopes of the younger Mauna Loa and the older Mauna Kea volcanoes (island of Hawaii, USA). Methods: After widespread forest decline (dieback), permanent plots were established in 1976 in 13 dieback and 13 non-dieback patches to monitor the population structure of M. polymorpha at ca. 5-yr intervals. Within each plot of 20 × 20 m, all trees with DBH >2.5 cm were individually tagged, measured and tree vigour assessed; regeneration was quantified in 16 systematically placed subplots of 3 × 5 m. Data collected in the subplots included the total number of M. polymorpha seedlings and saplings (five stem height classes). Here we analyse monitoring data from six time steps from 1976 to 2003 using repeated measures ANOVA to test specific predictions derived from the 1987 conceptual model. Results: Regeneration was significantly different between dieback and non-dieback plots. In dieback plots, the collapse in the 1970s was followed by a ‘sapling wave’ that by 2003 led to new cohort stands of M. polymorpha. In non-dieback stands, seedling emergence did not result in sapling waves over the same period. Instead, a ‘sapling gap’ (i.e. very few or no M. polymorpha saplings) prevailed as typical for mature stands. Canopy dieback in 1976, degree of recovery by 2003 and the number of living trees in 2003 were unrelated to substrate age. Conclusions: Population development of M. polymorpha supports the cohort dynamics model, which predicts rebuilding of the forest with the same canopy species after dieback. The lack of association with substrate age suggests that the long-term maintenance of cohort structure in M. polymorpha does not depend on volcanic disturbance but may be related to other environmental mechanisms, such as climate anomalies.