Abstract: The deformation data of retaining structures are considered to be one of the important indice to evaluate the safety and environmental impacts of excavation projects. However, because of the complexity of geological conditions and construction process, it is difficult to reflect the real stress condition with accurate value and shape of the deflection curve obtained by analytical solutions. Typical patterns of deflection curve are proposed based on the previous experience, soil mechanics and corresponding construction conditions. A method for pattern recognition of deflection curve based on the Bayesian probability theory is used to indicate the potential defects of retaining walls, and it has significance for improving the accuracy and efficiency of the informational construction. The proposed method is proved to be feasible by an actual project.