Modeling the terrestrial biosphere's carbon exchanges constitutes a key tool for investigation of the global carbon cycle, which has lead to the recent development of numerous terrestrial biosphere models. However, as demonstrated by recent intercomparison studies, results of plant carbon uptake, expressed as net primary productivity (NPP), still diverge to a large degree. Here, we address the question of uncertainty by conducting a series of sensitivity tests with a single, process-based model, the Biosphere Energy-Transfer Hydrology (BETHY) scheme. We calculate NPP globally for a standard model setup and various alternative model setups representing either changes in modeling strategy or approximate uncertainties of the most important model parameters. The results show that estimated uncertainties of many process parameters are still too large for reliable predictions of global NPP. The largest uncertainties come from plant respiration, photosynthesis and soil water storage. The surface radiation balance and day-to-day variations in weather, often not included into terrestrial vegetation models, are also found to contribute significantly to overall uncertainties, while stomatal behavior, the aerodynamic coupling of vegetation and atmosphere, and the choice of the vegetation map turn out to be relatively unimportant. A further comparison with field measurements of NPP suggests that such data are too unreliable for validating biosphere mod¿l predictions. We conclude that the inherent uncertainties in process-oriented biosphere modeling are able to explain the discrepancies that have occurred when comparing the results of different models. ¿ 2001 American Geophysical Union