Background/Question/Methods Understanding the drivers of forest community composition and stability constitutes a primary goal of forest ecology. Species' different demographic responses to the environment (i.e., the response of tree growth, mortality, and fecundity to the physical and biological environment) can scale up to explain community patterns such as coexistence and community stability. Predicting how forests might change in response to changing environments and species composition, therefore, requires models that can accurately capture both the demographic processes at work in forest populations and uncertainty in those processes. A new forest simulator, the scalable landscape, inference, and prediction (SLIP) simulator, uses hierarchical Bayesian estimates of whole life-history demography of trees and incorporates both process uncertainty and individual differences into forward simulation experiments. Process uncertainty describes unexplained variation in the mechanisms that drive demographic rates (e.g., the effect of unmeasured pathogens on growth). Individual differences indicate how each individual within a species can be more or less vigorous in their demographic response to the environment. Process uncertainty is species specific, but is associated with all individuals equally, whereas individual differences, also species-specific, are assigned to an individual for life. In a series of experiments using both a conceptual forest of five species and a replicated forest of 37 species, we tested the hypothesis that eliminating process error and individual differences would lead to the overestimation of demographic variance and therefore increase extinctions and reduce species coexistence. Results/Conclusions In a deterministic model (with no uncertainty modeled), extinctions were consistently higher than in runs that incorporated process uncertainty only (i.e., individual differences were added to process uncertainty), reflecting increased demographic stochasticity. Process error, which entered the model at each time step, decreased extinctions by buffering demographic responses to the environment. When individual differences were separated out from process error, demographic rates were buffered over and above the effects of process error alone by including ‘frailty' in forward simulations. In these runs, weaker individuals (those with lower than average individual responses) tended to die early, leaving more vigorous individuals in the population (just as in real populations). These vigorous individuals were better able to endure poor conditions and interspecific competition, maintaining higher species richness over model runs. As population viability analyses and forest response to climate change are becoming important focuses of simulation experiments, we suggest that the incorporation of both process error and individual differences are crucial to correctly anticipate future patterns of forest composition.