The effects of missing data (MD)and imputation methods(IM) in clus-ter analysis have been studied in, Silva (2005) and Silva et al. (2006), for somehierarchical classification methods and partition methods, in the case of variablesclustering. As in Silva et al (2006) the partition method is the following: we start byfinding a dissimilarity matrix between variables; a multidimensional scaling tech-nique (MDS)-PROXSCAL-provides components which are used as inputs in a k-means method. In this communication, when there are MD, we evaluate the effectof IM combined with the PROXSCAL MDS procedure (Commandeur and Heiser(1993)): for a data matrix with missing data;mimputations are realized;mdissim-ilarity matrices are then obtained from each imputed matrix; PROXSCAL withoutconstraints over thesemdissimilarity matrices provides components; k-means isperformed on these components and finally the partitions is compared with theoriginal one ie with the complete data by means of the Rand index as in Younessand Saporta (2004) and an affinity coefficient as in Sousa (2006). The simulationstudy consists in generating different patterns of partitions from twenty-five vari-ables following multinormal distributions. As in Silva (2005) data are deleted inincreasing proportions to create MD patterns and several IM are compare