Numerical simulation of staged construction of urban deep excavations is commonly used to estimate induced ground deformation in the support wall and at adjacent existing structures. During construction it is desirable to incorporate field observations from the early construction stages into the numerical simulation to obtain a more accurate estimate of anticipated ground deformations in later construction stages where the excavation level is deeper. A novel, powerful and systematic method to calibrate the constitutive model of the soil behavior directly from field measurements is developed. The autoprogressive method; a neural network based methodology that has been proposed by Ghaboussi and his co-workers, is applied to the modeling of staged construction for a deep braced excavation. Analysis results demonstrate the applicability of the proposed methodology. A discussion of the potential of this method is also presented.