A special approach to evaluation of separation performance is necessary for scattered response data caused by fluctuation of the composition of the feed alpha and operation variables. The fluctuations resulting from variation of alpha can be significantly reduced by using selected, insensitive to changes of alpha, separation plots and correlation lines aiming at the start and finish points of the upgrading plot. When data of two experimental series are scattered but not overlapped, parameters of the correlation lilies can be used for their comparison. When the data are scattered and overlapped, the use of appropriate confidence levels lines is recommended for comparison purpose. Then, the superiority of one series of data over another can be judged from the difference between their correlation and confidence lines. The correlation lines and non-overlapping confidence level lilies should be plotted for different feed compositions in insensitive-to-alpha plots since the result of comparison may depend on parameters, that is criterion, of the comparison. (C) 2004 Elsevier B.V. All rights reserved.