Various system identification methods have been introduced through many manners by using numerical techniques to validate a complicated structures which have been described in FEM by comparing measured modal data. The objection of this work is to propose a method to identify a structure by comparing measured modal data to the numerical FEM data. Identified structures will improve the accuracy to the numerical model by minimizing the differences between those two models. Numerical base-line model is constructed by using FEM and will be compared to perturbed model by solving inverse problem. Measured modal responses, which are eigenvalues and eigen vectors, will be applied to satisfy the equilibrium and to minimize the differences of modal responses between the original model and the perturbed model. In this study, a neural networks-based detection method using modal properties is presented as a method for the identification which can effectively consider the modeling errors. Due to lack of number of the sensors, degrees of freedom-based reduction method has been applied to restore full model. As neural network has been applied for identification method, efficiency in calculation time is expected to improve.