When the Mahalanobis-Taguchi system is used for recognition or discrimination,it is necessary to create a unit space, representing normal data, as a reference.In order to use the unit space to recognize multicomponent medicines fortesting purposes, however, the unit space should ideally be designed to toleratebatch-to-batch fluctuation of the medicine and variability of the analyticalmethods and equipment. It was thought that a unit space suitable for recognizingthe medicine under test could be obtained by adding tolerances for batch-tobatchfluctuation and analytic error. Unit spaces designed in this way weretested by Mahalanobis distance calculations. The threshold for pass-fail discriminationwas also studied, by calculating standard S/N ratios from the errorrate. Furthermore, the threshold value considering the loss was examined.