The RAM model performance in estimating SO2 concentrations in St. Louis, MO for 1976 has been discussed in several previous papers by the authors. In these studies the model estimates were compared to the observed concentrations of SO2 at 13 sites in the St. Louis metropolitan area. An analysis is made evaluating the RAM model performance in estimating the infrequently occurring high concentration values. Previous comparisons have been criticized for not assuring that the modeled high concentration values occurred during similar meteorological conditions as when the high concentration values occurred during similar meteorological conditions as when the high concentrations were observed. To address these concerns, the data are sorted by stability class and wind speed before the comparison statistics are computed. This allows a meaningful comparison of average concentrations and high values for each data grouping. Results are displayed graphically and provide insight into the variation of model performance with both stability class and wind speed. The resulting comparisons suggest that the RAM model concentration estimates may have a bias towards underestimating concentrations during very unstable conditions and overestimating concentration values during very stable conditions.