Downwelling surface solar radiation is an important input to ecosystem models, and global models require spatially extensive data sets that vary interannually to capture effects that potentially drive changes in ecosystem function. In this paper, I describe and compare solar radiation data sets from two representative sources, National Centers for Environmental Prediction (NCEP) reanalyses and Goddard Institute for Space Studies (GISS) calculations that included satellite observations of cloud properties. The CASA ecosystem model, which uses solar radiation and satellite-derived vegetation information, was run with the two solar radiation data sets to explore how differences affect estimated net primary production (NPP). GISS solar radiation matched ground-based observations better than NCEP solar radiation. Mean global NCEP solar radiation exceeded that from GISS by 16%, likely as a result of lower cloudiness within the NCEP reanalyses compared to satellite observations. Neither data set resulted in a significant trend over the study period (1984--2000). Locally, relative differences were up to 40% in the mean and 10% in the trend of solar radiation and NPP, and varied in sign across the globe. Because reanalysis solar radiation is only indirectly constrained by observations in contrast to the satellite-derived data, it is recommended that studies use the GISS solar radiation when possible.