The uncertainties in grassland ecosystem simulations caused by uncertainties in the parameters were studied using a theoretical five-variable grassland ecosystem model and a conditional nonlinear optimal perturbation (CNOP-P) method. Uncertainties in the parameters may originate in uncertainties in the observations and/or descriptions of the physical processes associated with the parameter, amongst other things. 32 model parameters that have physical meanings in the five-variable grassland ecosystem model were selected for use in numerical experiments. The results showed that when these parameters had the same degree of uncertainty, and the same optimization time, the combination of CNOP-Ps optimized for each parameter was different from the CNOP-P optimized for all 32 model parameters. The authors compared the grassland ecosystem simulations with the two types of parameter errors described above and with random parameter errors with the same degree of uncertainty as the optimized parameter errors. It was concluded that the CNOP-P for the 32 model parameters optimized at the same time led to the maximum uncertainty in the grassland ecosystem simulation, which was that the grassland ecosystem was either transformed into a desert ecosystem or another grassland ecosystem with more biomass. These results were independent of the size of the parameter uncertainties and the optimization time, and they show that nonlinear interactions between several parameters in the model are important to the uncertainties in the grassland ecosystem simulation. The results also imply that the CNOP-P method is a useful tool for assessing uncertainties in the grassland ecosystem simulation.