In order to discriminate three kinds of Chinese vinegars using an electronic nose (E-nose), an E-nose sensor array optimization method based on elimination transform with pivoting of Wilks Λ-statistic is put forward. With the help of the elimination transform some sensors with high discriminant ability are orderly selected step by step, and an ordered set of the selected sensors is simultaneously obtained by the selected sequence. Although the ordered set is not necessarily optimization array, the optimization array can be easily obtained by further exploring the discrimination results of sub-arrays which are composed by the first two, three, four and more sensors of the ordered set. In the discrimination of three kinds of vinegar samples, 9 sensors were selected in sequence and consisted of the ordered set using the method, and these vinegar samples could be well discriminated by principal component analysis (PCA). Furthermore, the first two sensors were selected and their PCA result was better than that of the 9 sensors. This illuminates that the array optimization method is very suitable and effective.