size of our study (50 patients) could lead to spurious results in factor analysis. However, no consensus exists onthe minimum sample size needed to perform reliable factor analysis; recommendations from the literature are varied, even contradictory (2). A recent simulation study determined the minimum necessary sample size for 180 different population conditions that varied in number of variables, number of factors, number of variables per factor, and level of communalities (4). To make these determinations, 37,160,000 sample correlation matrices were generated. The minimum required sample size varied from 40 to 1,300 depending on the condition, and the authors concluded that "attempts to provide an absolute minimum necessary sample size are probably unrealistic."In our article, we also discussed the problem of the minimal sample size needed for factor analysis, and obviously a larger sample size leads to more reliable results. However, our conclusions, based on Spearman's correlation coefficient results, are more reliable because of the rather small confidence intervals (CIs) for these coefficients obtained with 50 patients. Therefore, the unexpected strength of the correlation (0.57; 95% CI 0.35-0.73) between the physical component score (PCS) and mental component score (MCS) in our sample suggests that the a priori stratification of the 8 subscales of the SF-36 might not be adapted in certain situations, and that the factorial structure of the scale may depend on the type of pathology. This hypothesis is supported by our recent study involving a sample of 2,540 patients with knee osteoarthritis and 1,593 with hip osteoarthritis, showing that the PCS and MCS a priori structure is not confirmed by factor analysis. The factorial structure of the 8 subscales obtained from patients with knee or hip osteoarthritis was similar, but it differed from that obtained from patients with SSc (4).It is widely accepted that in order to establish the construct validity of a scale, several studies with convergent results are needed (5). Therefore, comparing the factorial structure and the convergent and divergent validities of the SF-36 obtained in different samples of SSc patients might be an exciting challenge.