AbstractWe used the Terrestrial Ecosystem Model (TEM) to investigate how alternative input data sets of climate (temperature/precipitation), solar radiation, and soil texture affect estimates of net primary productivity (NPP) for the conterminous United States. At the continental resolution, the climates of Cramer and Leemans (C&L) and of the Vegetation/ Ecosystem Modelling and Analysis Project (VEMAP) represent cooler and drier conditions for the United States in comparison to the Legates and Willmott (L&W) climate, and cause 5.2% and 2.3% lower estimates of NPP. Solar radiation derived from C&L and given in VEMAP is 32% and 60% higher than the solar radiation data derived from Hahn cloudiness. These differences cause ∼ 8% and 10% lower NPP because of radiation‐induced water stress. In comparison to the FAO/CSRC soil texture, which represents most biomes with loam soils, the soil textures are finer (more silt and clay) in the Zobler and VEMAP data sets. The use of VEMAP soil textures instead of FAO/CSRC soil textures causes ∼ 3% higher NPP because enhanced volumetric soil moisture causes higher rates of nitrogen cycling, but use of the Zobler soil textures has little effect. In general, NPP estimates of TEM are more sensitive to alternative data sets at the biome and grid cell resolutions than at the continental resolution. At all spatial resolutions, the sensitivity of NPP estimates represents the impact of uncertainty among the alternative data sets we used in this study. The reduction of uncertainty in input data sets is required to improve the spatial resolution of NPP estimates by process‐based ecosystem models, and is especially important for improving assessments of the regional impacts of global