Southern blot analysis of terminal restriction fragments (TRFs) is the standard method for quantitative examination of telomere length distributions. Since TRFs contain a subtelomeric component, central parameters of the TRF distribution n(L) such as the arithmetic mean (M) or the median (Me) cannot be derived directly from Southern blot data, i.e. from the optical density distribution OD(L). Several estimates have been applied instead; the seeming arithmetic mean A, the "center of mass" C, and the positions of maximal (P) and half-maximal optical density (P(1/2)). We show that C> A> M for any non-truncated distributions n(L), and P> M> P1/2 for any symmetrical unimodal n(L). Symmetric appearance on a Southern blot, however, suggests positive skewness of n(L). Thus, a lognormal form of n(L) may be considered. Then, C> A> M> P=Me> P(1/2). Alternatively, a Weibull distribution may be assumed. The latter is compatible with negative feedback-regulation of the telomere lengths. Using the maximum likelihood method we compare these distributions with FISH-data on telomere lengths in different cell types. The fit of the lognormal distribution is clearly superior. Lognormal genesis may relate to telomere breakage and recombination. Truncation of the upper end of the TRF distribution is possible due to Southern blot artifacts. Thereby, the order of the estimates may change to P> C> A. Having minimal sensitivity to truncation, P seems to be the optimal choice. however, the variability of P is high since peakedness of OD(L) and DNA length resolution are inversely related.