Abstract Serious problems arise in the maintenance of reliability in large data bases, since it becomes difficult to verify incoming data manually. This article considers the case in which the base consists of data vectors following a multivariate normal distribution. Five screening procedures are proposed—a “one-at-a-time” test, the standard χ2 test, and three statistics derived from principal component analysis. From analysis of a practical example, it emerges that the statistics derived from principal component analysis have superior performance.