Recently, statistical analysis of molecular similarity matrices has been applied to the quantitative structure-activity relationship (QSAR) analysis of a number of molecular series. This paper addresses a number of methodological issues relative to the similarity matrices. A series of halogenated aliphatic hydrocarbons, for which the mutation (aneuploidy) induction ability had previously been determined, was used as test bench. The chemical information carried by the similarity matrices was shown to overlap to a considerable extent the information carried by the classical descriptors (physical chemical and quantum mechanical parameters). A good QSAR was obtained on the basis of the similarity matrices, in analogy with that obtained with the classical descriptors; however, the similarity matrices neither complemented the classical descriptors nor were able to improve on their performance. The effect of the compound's spatial orientation on the similarity values was also investigated.