The assessment of skeletal maturity is crucial for the analysis of growth disorders and plays an important role in paediatrics. For this reason, several methods have been developed for estimating skeletal maturity. Among them, the Tanner and Whitehouse method (TW2), which is based on the analysis of hand radiographs, is usually considered the most accurate and reliable. Nevertheless, TW2 is applied only in a small fraction of cases, due to its complexity and long examination times. Thus, the development of automated systems which reliably implement this method is highly desirable. However, major difficulties have been found in the development of computer-based systems for the assessment of skeletal maturity. In particular the extraction of the bones of interest has proved to be extremely challenging. In this paper, we propose a system architecture for the implementation of the TW2 method, which is based on artificial neural networks. For each bone considered, the maturation stage is determined by means of a two-step process which first locates the position of the bone in the radiograph and then analyzes the bone shape. Experimental results obtained with our implementation of the carpal version of TW2 are in good agreement with those provided by trained observers.