AbstractA new technique for representative subset selection is presented. The advocated method selects unambiguously the most important objects among the calibration set and uses this subset for the model development without significant deterioration in the predictive ability. The method is called boundary subset selection and it is an inherent part of the Simple Interval Calculation (SIC) approach. SIC is a method for linear modeling, which is based on the assumption of error boundedness. The primary SIC consequence is an object status classification (OSClas) that reveals the most influential objects and also designates the most stable and reliable ones. The OSClas is used as the main tool for representative subset selection. The presented results are compared with widely used Kennard–Stone algorithm and D‐optimal design procedure employing three real‐world examples. Copyright © 2008 John Wiley & Sons, Ltd.