Natural frequencies are appropriate damage-sensitive features to detect structural changes in cables and prestressing tendons as well as in the support displacements of large engineering structures, such as bridges. However, environmental changes often cause variations in the frequencies that are more pronounced than changes due to small structural damages, making it imperative to apply reliable machine learning approaches for the removal of environmental conditions. This case study sets out to demonstrate that small changes in the structural mass and failing bolts in a 30 m steel-concrete composite, the Munich Test Bridge, can be detected despite large temperature variations. During the 21 daylong measurement campaign, no traffic loads are applied to the bridge but the ambient temperature ranged between -10°C and 20°C. An artificial neural network is applied to remove the effect of ambient temperatures on the measured frequencies in a separate preprocessing step. Ultimately, it is demonstrated that, after the removal of temperature effects, the distribution of natural frequencies can be approximated through a normal distribution, paving the way for more advanced analysis techniques that require features to be normally distributed.