Characterizing reservoir quality, identifying the main rock types, and predicting their spatial variations are a challenge. Supervised pattern-recognition methods are used as discriminant analysis of these parameters. However, the uncertainties of the measurement arrays are not considered, which may cause misinterpretations. A methodology was developed that is an extension of the standard parametric approach to discriminant analysis. The resulting reservoir-quality model is less precise but more realistic by taking into account all data and associated uncertainties.