Journal of Cereal Science 14 (1991) 95-104 Principal Component Analysis - an Efficient Tool for Selection of Wheat Samples with Wide Variation in Properties E. WESTERLUND, R. ANDERSSON, M. HAMALAINEN and P. AMAN Swedish University of Agricultural Sciences, Department of Chemistry, Box 7015, S-750 07 Uppsala, Sweden Received 10 July 1990 Twenty-three variables were studied that described the chemical composition and baking properties of 100 samples of spring and winter wheat. Sixteen of these samples were selected with the aid of principal component analysis (PCA) in such a manner that much of the variation in all the parameters was retained. The selection procedure preserved a larger part of the variation in the original material than selection by random sampling. This was deduced from comparison of ranges recovered with respect to each variable and selection procedure. The principles of PCA are general and the procedure developed can therefore be recommended as one means of solving sample selection problems. Introduction Selection of a number of subsampies from a larger sample set in a way that retains as much as possible of the variability present in the original sample set constitutes an important problem in applied science. One way to select samples is to use random sampling\ but this may produce a high error in subsequent analysis due to unfortunate selection of non-representative samples. Performing the selection in a systematic manner is generally more attractive, therefore, depending on the particular aim of the investigation 1 . When the number of samples and the number of sample variables are large, however, systematic selection based on visual comparison of sample values may become difficult. The baking properties of 100 wheat samples are being studied in this laboratory by evaluating a multitude of chemical and physical parameters. A detailed characterization of soluble fibre components, which are considered to have important baking properties2, became necessary. As a consequence, the original number of wheat samples had to be reduced by selection because of the extensive analytical work that would otherwise be required. A suitable selection procedure was developed using principal component analysis (peA), which is a mathematical multivariate method 3 • 4 • Multivariate methods have been used previously to select samples for multivariate calibration of near-infrared spectrophotometers·- 7 • The aim of the present paper is to show that peA can be used to reduce the number 0753-5210/91/040095 + 10 $03.00/0 © 1991 Academic Press Limited