This paper presents a damage detection method using a combination of global (changes in natural frequencies) and local (displacement mode shapes) vibration-based analysis data as input in neural networks for location and severity prediction of structural damage. The necessary features for damage detection have been selected by performing sensitivity analyses. In order to check the robustness of the input used in the analysis and to simulate the numberical errors, artificial random noise has been generated numerically and added to noise-free data during the training of the neural networks. Furthermore, a modified back-propagation neural networks with the dynamic steepest descent (DSD) algorithm as training algorithm is used to improve the training efficiency. With those as basis, the neural networks can assess damage condition (locaiton and severity) with very good accuracy.