In Australia, the Full Carbon Accounting Model (FullCAM) is used by the Australian Government for international reporting of greenhouse gas emissions and to predict carbon stock changes for carbon abatement projects. Consequently, over the last 20 years, it has been routinely applied at continental, regional and local scales, and has been subject to on-going development to improve its accuracy and representativeness. Given its importance, a sensitivity analysis could facilitate our understanding of model behaviour and aid the planning of future work and data collection. In particular, the sensitivity of a given model parameter is often context dependent such that it depends on the level of other parameters or variables.Key FullCAM parameters have generally been calibrated using data from empirical studies, with differing levels of confidence based on the sample size and data quality. The objective of this study was to apply a sensitivity analysis to examine (i) the sensitivity of FullCAM carbon stock outputs to its parameters and inputs, and how the sensitivity indices compare with the sample sizes used to calculate the respective parameters, (ii) the context dependency in terms of how the sensitivity varies with age, potential maximum biomass (a key FullCAM parameter), and disturbance severity or type, and (iii) to identify the implications for further development of FullCAM for woody vegetation systems.Of the 67 parameters tested, FullCAM carbon stock outputs were most sensitive to background mortality rates, age, potential maximum biomass, climate variability, age of maximum growth, decomposition or turnover parameters, and stand structure (regenerating or mature). The context dependency of the sensitivity analysis followed a consistent pattern depending on two main conditions, such that the sensitivity was higher when (i) the carbon stock was large and (ii) the parameter had a strong influence on that carbon stock. Several strong context dependencies occurred because the size of carbon stocks often vary through space and time within an ecosystem, and different processes (i.e. parameters) are more important at different times and locations. There was a strong context dependency in relation to age and potential maximum biomass, and FullCAM simulations indicated an interaction between fire disturbances and vegetation types. Lastly, background mortality was one of the least available inputs, but one to which FullCAM outputs were very sensitive. FullCAM was applied to show that the long-term contributions of low rates of background mortality to standing dead and debris C stocks were often as large as those from frequent (every 30 years) and intense harvesting and fire events because of the regular (annual) contribution of mortality and its accumulating influence of C stocks.