A detailed 3-D Finite Element (FE) model was developed for a cable-stayed bridge, and natural frequencies were compared with those from ambient vibration tests. The model was updated using a systematic Manual Tuning technique to improve the match, and the final results are comparable with those using automated Model Updating on similar bridges. The Manual Tuning method investigated the match between frequencies for torsional and vertical modes separately, which aided identification of modelling deficiencies. These can be minimised using a detailed representation of the mass distribution across the width of the bridge deck, and a consistent mass matrix, and by preceding the modal analysis with a geometric non-linear static analysis. An advantage of Manual Tuning is that the mesh configuration and element types can be modified, as well as the structural parameters, providing a closer representation of the structure. Further refinement of the parameters can be achieved with subsequent Model Updating. Each bridge cable was represented by a single truss element, with the elastic modulus modified to account for sag. For the bridge studied, it was found that the cable tension used to calculate the modulus could vary significantly but still give reasonable results, if realistic initial estimates were made. The natural frequency ratio of the first vertical and torsional deck modes varied significantly from the initial to the final model. The variation of this ratio, used to establish the potential for classical flutter, highlighted the importance of accurate FE modelling for design. (c) 2006 Elsevier Ltd. All rights reserved.