Two process-based models were used to identify the environmental variables limiting productivity in a pristine, mature forest dominated by rimu (Dacrydium cupressinum Sol. ex Lamb.) trees in South Westland, New Zealand. A model of canopy net carbon uptake, incorporating routines for radiation interception, photosynthesis and water balance was used to determine a value for quantum efficiency when climate variables were not limiting. The annual net carbon uptake by the canopy was estimated to be 1.1 kg C m(-2) and the quantum efficiency 22.6 mmol mol quanta(-1). This value of quantum efficiency, combined with other parameters obtainable from the literature, was then used in a model of forest productivity (3-PG), to simulate changes in net productivity and the allocation of carbon to tree components. The model was adjusted to match a measured stem increment of 10.6 Mg ha(-1) over a period of 13 years. To achieve this while maintaining a low, but stable value for leaf area index, it was necessary to set the site fertility rating very low and select high values for the parameters describing the proportional allocation of total carbon to roots. This approach highlighted nutrient availability as the principal constraint on productivity for the ecosystem and identified critical measurements that will be necessary for using the model to predict the effects of climate change on carbon sequestration. The low rates of carbon uptake and productivity are consistent with the low nutrient supply available from the highly leached, acid soils, most likely attributable to frequent saturation and a very shallow aerobic zone.