A generalized terrestrial ecosystem process model, BIOME‐BGC (for BIOME BioGeoChemical Cycles), was used to simulate the global fluxes of CO2 resulting from photosynthesis, autotrophic respiration, and heterotrophic respiration. Daily meteorological data for the year 1987, gridded to 1° by 1°, were used to drive the model simulations. From the maximum value of the normalized difference vegetation index (NDVI) for 1987, the leaf area index for each grid cell was computed. Global NPP was estimated to be 52 Pg C, and global Rh was estimated to be 66 Pg C. Model predictions of the stable carbon isotopic ratio 13C/12C for C3 and C4 vegetation were in accord with values published in the literature, suggesting that our computations of total net photosynthesis, and thus NPP, are more reliable than Rh. For each grid cell, daily Rh was adjusted so that the annual total was equal to annual NPP, and the resulting net carbon fluxes were used as inputs to a three‐dimensional atmospheric transport model (TM2) using wind data from 1987. We compared the spatial and seasonal patterns of NPP with a diagnostic NDVI model, where NPP was derived from biweekly NDVI data and Rh was tuned to fit atmospheric CO2 observations from three northern stations. To an encouraging degree, predictions from the BIOME‐BGC model agreed in phase and amplitude with observed atmospheric CO2 concentrations for 20° to 55°N, the zone in which the most complete data on ecosystem processes and meteorological input data are available. However, in the tropics and high northern latitudes, disagreements between simulated and measured CO2 concentrations indicated areas where the model could be improved. We present here a methodology by which terrestrial ecosystem models can be tested globally, not by comparisons to homogeneous‐plot data, but by seasonal and spatial consistency with a diagnostic NDVI model and atmospheric CO2 observations.