Background: The aim of this study was to develop a child-specific classification system for long bone fractures and to examine its reliability and validity on the basis of a prospective multicentre study. Methods: Using the sequentially developed classification system, three samples of between 30 and 185 paediatric limb fractures from a pool of 2308 fractures documented in two multicenter studies were analysed in a blinded fashion by eight orthopaedic surgeons, on a total of 5 occasions. Intra- and interobserver reliability and accuracy were calculated. Results: The reliability improved with successive simplification of the classification. The final version resulted in an overall interobserver agreement of kappa=0.71 with no significant difference between experienced and less experienced raters. Conclusions: In conclusion, the evaluation of the newly proposed classification system resulted in a reliable and routinely applicable system, for which training in its proper use may further improve the reliability. It can be recommended as a useful tool for clinical practice and offers the option for developing treatment recommendations and outcome predictions in the future.