­In this study the proper orthogonal decomposition method is utilised as a model reduction technique in crack size estimation in a cracked plate under traction problem. The idea is to create a reduced model based on the results issued from finite element method, thus the crack size parameter is directly related to the boundary displacement obtained from the boundary nodes considered as sensor points. The inverse investigation is run using a genetic algorithm to minimization the error function expressed as the difference between data caused by the crack proposed by genetic algorithm in every individual and the one measured at the actual crack identity. The reduced model is validated by comparing the estimated structural response with the corresponding results from the finite element model. The effectiveness of the approach related to the used number of sensors is presented. Finally the stability of the method against uncertainty is tested by introducing different levels of white noise to the reference data.