An investigation into the reproducibility of TW2 skeletal age assessments was carried out using three observers. The first, G.B., had learned the technique without expert advice, while the second, R.H.W., was one of the originators of the technique. The third, N.C., was an experienced rater orginally trained by R.H.W. G.B. and R.H.W. were compared using a standard set of 112 X-rays arranged in a random order. G.B. was compared to N.C. using a random sample of 50 X-rays from the Leuven Growth Study of Belgian Boys. The percentage agreement in rating all 20 bones was just over 83% in both analyses. Higher percentage agreements were found in both analyses for the round bones (88% and 85.1% respectively) as opposed to the radius, ulna and short bones (81.3% 82.1% respectively). Significant differences were found in mean bone-age between G.B. and R.H.W. and between G.B. and N.C. for TW220 and carpal bone ages. No significant differences in mean bone-ages were found for RUS bone-ages. The reproducibility of the self-taught assessor was found to be as good as that of assessors trained by the originators of the technique or experienced raters.