A sensitivity analysis on the impact of uncertainty in tree mortality on the predictions of an individual-based spatially explicit forest dynamics model (SELVA) was performed. The model was developed to investigate the impact of felling (logging) on the demography and structure of tree populations in tropical forests (French Guiana). This study addressed questions about (1) the relative impact on model predictions of uncertainty in mortality processes at different stages of tree development; (2) the interactions between the mortality processes and different felling regimes; and (3) the impact of different felling regimes on the demography and structure of tree populations, taking account of answers to (1) and (2). A global approach of sensitivity analysis based on the decomposition of the output variance was applied. Based on prior knowledge about model uncertainties, mortality processes at the stages germinated seed, standing juvenile, standing adult, and tree-fall were focused as input factors. The input factors were multivariate mortality sub-models involving several parameters with no explicit biological meaning. Thus, an approach based on confidence ellipses of parameter estimates was used to vary input factors homogenously, so that the impact of different input factors on a given model output could be compared. As outputs, the numbers of living, dead and recruited trees, and the tree diameter structure were analysed over 336 years of simulation. An additional local sensitivity analysis provided deeper insights into the relationships between model input and output. The results showed that standing juvenile mortality was the largest source of uncertainty, ahead of standing adult mortality, germinated seed mortality and tree-fall. Moreover, mainly standing juvenile mortality interacted with the felling regime, resulting in changes of the diameter structure of the studied tree population (Dicorynia guianensis Amshoff, Caesalpiniaceae). Felling all trees =60 cm dia