Historical anthropology (physical) is becoming involved in large scale analysis: geographical and chronological. In view of this extension in scope-and in morphological variability-three current distance coefficients are examined as to their mathematical and taxonomical properties. Work was done on the data collected by Howells (1973). Penrose distances proved to be methodologically unsatisfactory (although they can produce useable results under some conditions). The size component revealed to be very sensitive (unstable) with regard to the variates choosen ; its importance with respect to shape decreases as a fonction of the number of variates. The breakdown of total distance into size and shape is arbitrary and almost depleted of morphological meaning. Generalized distance ( Mahalanobis) : if conditions for correct application are to be respected, D^{2 can only be used in problems of low taxonomie level (population, race, but not species). - D2 and centroid distances in discriminant space are equivalent only if discriminant analysis is carried out according to Fisher's algorithm. Centroid distances in principal component space appear to be a method of general interest and purpose since they result from a "neutral" maximisation procedure and are not affected by restrictive conditions as to their application. As for the breakdown of the total metric information into size and shape components, it is demonstrated that these two categories are surprisingly independant. More emphasis should be given to metrical assessment of shape (stereometry). A method for splitting total distance into size and shape components is presented. Substantial refinement of interpretation of anthropometrical data can be expected.}