Abstract The Terrestrial Vegetation Model (TVM) of the Integrated Model to Assess the Greenhouse Effect ( image 2.0) determines potential land cover for natural ecosystems and potential productivity for agrosystems. A model analysis of this TVM using Monte Carlo simulations is described for the eighteen Costa Rican (0.5° × 0.5° latitude-longitude) image 2.0 grid cells. It is demonstrated that the TVM is not overly sensitive to climate input variability. Much observed output variability is related to model parameters. These model parameter related effects can strongly determine the intra/inter grid variability for Costa Rican potential yields. The model analysis indicates that more model refinement is required to make any effort to improve data quality effective. It is therefore proposed to remove as many model criteria as possible or to replace them by more gradual criteria based on crop ecophysiological characteristics. A first tentative comparison of maize, rice and pulses yield potentials and their distribution in Costa Rica during 1973 and 1984, suggests that the assumption that the two are related, as assumed in the Land Cover Model (LCM) in image 2.0, has some validity.