Steele's (1970) regression method for estimating femur and tibia length from fragmentary bones is tested on a sample of complete femora (female N = 26; male N = 33) and tibiae (female N = 16; male N = 22) from a number of European Mesolithic and Neolithic sites. Over half of the regression equations given by Steele for predicting maximum length of the bone from the length(s) of one or more of its constituent segments are shown to produce inaccurate predictions in this test sample. However, a closer evaluation of these results, including calculation of regression equations for the test sample itself, reveals that this inaccuracy does not derive from any inherent flaw in Steele's method. Rather, it is shown that differential distribution of maximum bone length among the various bone segments as defined by Steele may occur due to variation in muscular activity pattern and intensity. This argues for the retention of Steele's basic method, with care being taken to match closely the activity pattern typical of the sample from which regression equations are derived with that of the population to which the equations are to be applied. The equations calculated in this study thus are provided for use where deemed appropriate. © 1992 Wiley-Liss, Inc.