To investigate the serviceability reliability of a braced excavation system, a means has to be established for estimating wall deflection in the braced excavation. Traditionally, the wall deflection and ground settlement are "predicted" using finite element method (FEM) and/or other simplified methods. In this study, artificial neural networks (ANNs) are developed as an alternative of estimating the maximum wall deflection. The developed ANNs are validated with the collected 8 case histories of deep excavation that were not used in the development of these ANNs. Copyright ASCE 2006.