This paper aims at describing different analysis steps and findings regarding vibration measurement performed on a very long viaduct (about 20 km) with railway traffic. The initial goal of the measuring campaign was to identify modal properties from a given section of the viaduct to enhance possible structural health monitoring (SHM) techniques to be applied in the near future. A preliminary postprocess has shown high fluctuations in fundamental modal parameters like the resonant frequency of the first bending mode of the viaduct, while these results were expected to be consistent with time, regardless of the train speed. Advanced processing techniques based on wavelets have been used to clearly identify the behavior of the potential modes, showing a high sensitivity to train speed and potential non-linear effects. Given that the study focuses on a long viaduct, neighboring bridge span effects must also be taken into account. To complement the previous time-frequency analysis, a covariance-driven stochastic subspace identification method (SSI-COV) was also applied to the experimental data. Limits regarding this modal extraction method when dealing with non-linear behavior and time varying modal properties are discussed and evaluated with respect to data obtained with wavelet transforms. As a second step, sensitivity of the results based on pass-by speed and restricted time windows are presented and discussed. Furthermore, statistical data obtained from more than 120 measured trains are shown, with emphasis on recommended methods to reduce datasets which might lead to increased consistency in the evaluation of modal parameters.