Abstract: Removal of allometric variance (RAV) based upon regression of character ratios (indices) against body size may considerably improve species discrimination in ants when only single or very few characters are available or considered. Thisclear advantage of Rav is lost in multi-character discriminant analyses (DA) in the moderately allometric ant generaHypoponera, Temnothorax, Cardiocondyla, Myrmica, Lasius, and Formica. In order to asses if this result is repeated instrongly allometric organisms, two sister species of the ant genus Camponotus, C. ligniperda (Latreille, 1802) and C.herculeanus (Linnaeus, 1758), were investigated. Several DAs using primary absolute data, simple ratios against bodysize, log-transformed and RAV-corrected data were run. On the single-character level, the interspecific overlap ratio wasreduced from 71.2 % in primary ratios to 21.0 % in RAV-corrected ratios. In comparison to any other data type tested,RAV-corrected data showed an improved performance also in multi-character DAs. The advantage of RAV-correctedagainst log-transformed data is significant with p < 0.005 for combinations between two and five characters but is verysmall in a six-character DA. Rav is expected to constantly loose its benefits with increasing number of characterscomputed. This does not exclude residual advantages when the species considered are extremely similar and strongly allometric. Apart from their variable contribution to species discrimination, RAV-transformed data have the constant advantage of exposing diagnostic characters and of unmasking pseudo-characters, i.e., to show interspecific differencesrather independent from environmental factors that may have influenced worker body size. For this reason, Rav is always recommended in the taxonomic practice. Procedures of the Rav technique are explained in more detail and working routines of how to treat monophasic and diphasic allometries are presented. It is recommended to start any RAVanalysis with a graphical analysis. Predictions that measurement error and natural variation of Cs (cephalic size) significantly reduce regression slopes in index-formulae of the type Y / Cs = a * Cs + b were confirmed by simulations. Thisslope reduction, however, is meaningless because the second step of the Rav technique completely compensates for this.Predictions of absolute measurements Y by the index-formula and by logarithmic descriptions of the type log Y = log b + a* log x are nearly equal in 13 Temnothorax and Formica species with monophasic allometries.