Four mathematical models simulated decay of two litter types of contrasting quality over a 2-year period at four sites in North America. The litter types were Drypetes glauca and Triticum aestivum, representing litter with high and low nitrogen:lignin ratios, respectively. The field sites were an Arctic tussock tundra (Alaska, United States), a warm desert (New Mexico, United States), a temperate deciduous forest (New York, United States) and a tropical rain forest (Puerto Rico). Models captured the overall patterns of site and litter quality controls on decomposition; both simulated and observed mass losses were higher in warm, moist environments (both forests) than in cold (tundra) or dry sites (desert), and simulated and observed decay was more rapid for Drypetes than Triticum. However, predictions tended to underestimate litter mass loss in the tropical forest and overestimate decay in the desert and tundra, suggesting that site controls in model formulations require refinement for use under such a broad range of conditions. Also, predicted nitrogen content of litter residues was lower than observed in Drypetes litter and higher than observed for Triticum. Thus mechanisms describing loss of nitrogen from high-quality litter and nitrogen immobilization by low-quality litter were not captured by model structure. Individual model behaviors revealed different sensitivities to controlling factors that were related to differences in model formulation. As these models represent working hypotheses regarding the process of litter decay, results emphasize the need for greater resolution of climate and litter quality controls. Results also demonstrate the need for finer resolution of the relationships between carbon and nitrogen dynamics during decomposition. ¿ 1999 American Geophysical Union