The likelihood of simulated rainfall above a specified threshold being observed is evaluated as a function of the amounts predicted by two mesoscale models. Evaluations are performed for 20 warm-season mesoscale convective system events over the upper Midwest of the United States. Simulations were performed using 10-km versions of the National Centers for Environmental Prediction Eta Model and the Weather Research and Forecasting (WRF) model, with two different convective parameterizations tested in both models. It was found that, despite large differences in the biases of these different models and configurations, a robust relationship was present whereby a substantial increase in the likelihood of observed rainfall exceeding a specified threshold occurred in areas where the model runs forecast higher rainfall amounts. Rainfall was found to be less likely to occur in those areas where the models indicated no rainfall than it was elsewhere in the domain; it was more likely to occur in those regions where rainfall was predicted, especially where the predicted rainfall amounts were largest. The probability of rainfall relative-operating-characteristic and relative-operating-level curves showed that probabilistic forecasts determined from quantitative precipitation forecast values had skill comparable to the skill obtained using more traditional methods in which probabilities are based on the fraction of ensemble members experiencing rainfall. When the entire sample of cases was broken into training and test sets, the probability forecasts of the test sets evidenced good reliability. The relationship noted should provide some additional guidelines to operational forecasters. The results imply that the tested models are better able to indicate the regions where atmospheric processes are most favorable for convective rainfall (where the models generate enhanced amounts) than they are able to predict accurately the rainfall amounts that will be observed.