An extensive assessment was made of nine common similarity measures on
the basis of their discriminative ability and bias in weighting of di
fferent types of variation in species abundance between samples, by us
ing sets of river macroinvertebrate samples. Seven ag glomerative hier
archical clustering methods were applied to these measures. Successful
site discrimination is defined as grouping of all replicate samples f
rom a particular site. Major differences existed in the discriminative
ability between the similarity measures. The measures overweight some
types of variation and underweight the others to varying extents, whi
ch is closely related to their discriminative ability. A new dissimila
rity-similarity measure was devised to respond to all types of variati
on with minimum bias. This measure yielded a higher percentage of corr
ect site discrimination than the others tested. Use of an extra data s
et confirmed the superior performance of the new measure and also indi
cated that it could discriminate between sites of different water qual
ity.