The quality of multicomponent samples from one or several groups of samples can be monitored by a pattern recognition method . The method is based on profiles of sample quality, which are obtained by means of a multicomponent analytical technique (e .g ., ultraviolet spectroscopy or chromatography), and data reduction is done with the aid of fuzzy set theory . The advantages of the method in cases of overlapping and non-additive signals are outlined for quality control of analgesic tablets by ultraviolet spectroscopy . Its performance in the case of highly uncertain data patterns is demonstrated for classification of protein samples by chromatography . Spectroscopic and chromatographic methods are used to characterize multicomponent samples with respect to chemical components and their concentrations . Apart from the need to report accurate component concentrations, the analyses are frequently applied to ensure the quality of a final product (e .g ., by comparing the metal pattern of a steel sample with that of a certified standard) or to classify a complex sample by means of its chromatographic profile used as a "fingerprint" . The required data-processing schemes are essentially methods of pattern recognition based on supervised learning techniques [1] . However, the commonly used methods, such as the k-nearest neighbour, the measurement of nearness to the centre of gravity, or the linear learning machine, are not very suitable for solving the problem envisaged because at best only a gross description of the class shape is possible, and the ratio between objects (samples) and features (sensors) is far from being about 3 :1 .
Individual class-modelling techniques are of more use, as was shown by Wold et al .
[2] who used the SIMCA approach for fingerprinting human brain tissues profiled by means of gas chromatography . The SIMCA method is based on principal components analysis [3] ; any known class is associated with an individual mathematical model (e .g ., a plane or cube) which is then used to decide whether a sample pattern falls within or close to the class . Despite its good performance in many examples [1-3], the SIMCA method has