The original or revised methods for prediction of adult stature of Tanner-Whitehouse (TW), Roche-Wainer-Thissen (RWT), and Bayley-Pinneau (BP) were validated in a mixed longitudinal study of Turkish girls, 9-14 years of age. Comparison of 727 estimations of 102 girls showed that the RWT'93 revision is an improvement over the original method by reducing the standard deviation, and absolute median and mean of residuals. The TW'83 (3v) method, by dropping the variable midparent stature, was reduced in accuracy before age 12. The addition of the previous annual increment in stature improved the predictions, while further addition of the increment in skeletal age did not. In relation to overall accuracy and reliability, BP (0.75) and TW'75 (0.73) had the highest R(2) values, while RWT'75 had the lowest (0.55). Regarding individual consistency of multiple predictions, the RWT methods ranked best and TW'83 relatively poorest. Systematic errors found in Turkish girls were similar to those found in French, Swiss, and Finnish girls. (C) 1995 Wiley-Liss, Inc.