For moderate or high conversion copolymerization data three sorts of peculiar data points are discussed which are disadvantageous in determining the relative reactivity ratios: (a) copolymer contains an unknown quantity of homopolymer because the comonomer was fully consumed before the end of the reaction, (b) estimation error of the conversion is relatively high, (c) estimation error of the copolymer composition is very high. For the numerical integration procedure according to the terminal model a selection is proposed to eliminate such data points from calculations which are suspected of being uncertain. Furthermore, a separate interpolative procedure is proposed which permits a final separation of points of type (a) from points of type (b) and (c). (C) 1997 Elsevier Science Ltd.