In application of Monte Carlo sensitivity analysis to an ecosystems model in cattle grazing grassland, robustness of the method was investigated. The model used in this study was a compartment model containing 9 state variables and 66 model parameters. First, the effects of number of iterations (i.e., number of model runs when model parameters were changed) on sensitivity conclusions were examined. The sensitivity conclusions for predictions of all state variables on day 120 were consistent irrespective of number of iterations, but those for predictions of 3 state variables on day 180 were different in 100 iterations from other iterations. The results indicated that consistent sensitivity conclusions require at least 500 iterations. Secondly, effects of the magnitude of model parameters on sensitivity conclusions were examined. On day 120, the sensitivity conclusions for predictions of all state variables were consistent, but different sensitivity conclusions for predictions of 3 state variables were drawn. The above two results suggested that the three predictions may be inconstant against number of iterations and change of model parameters. Thirdly, the sensitivity conclusions of present method were compared with those in conventional sensitivity analysis. The sensitivity conclusions between the two methods had almost perfect correspondence on day 120. Nevertheless, the sensitivity of a few model parameters on day 180 varied by the methods. The model parameters showing the different results were common in that they were used as asymptotic (maximum) values of logistic curves in the ecosystems model.