A method is developed to classify QSARs or PPRs according to their validity, with the validity given in terms of applicability and accuracy. As a first step, test sets of chemicals are classified with respect to the presence or non-presence of 15 structural elements by use of a bit-pattern. To account for the accuracy, deterministic outliers are defined based on variable confidence intervals. A quality function is introduced that relates the number of outliers to the length of that interval. Characteristic features of this quality function may be used to classify a QSAR or PPR together with quantities describing the test set.