Three to 16 animals were necessary to detect a difference of 50-20% in total collagen contents in Longissimus thoracis muscle, between two experimental groups of cattle, using the facilities available in our laboratory. The inter-assay variability was higher for insoluble collagen content than for total collagen content. Consequently differences between two groups are determined less accurately for insoluble collagen content than for total collagen content. This may be partly compensated by repeated measurements (minimum 4) to detect a difference of 40% or even higher. The low reproducibility of type I and type III collagen contents determined by electrophoretic analysis is due to uncontrollable technical parameters. Normalisation of the results is thus necessary for accurate quantification. The recommended procedure is to use total protein content detected and quantified on the gels as a covariable in the statistical model. This procedure induces a 15% reduction in intra-subject variability, and hence in sample size required to detect significant differences.