Within a research project dealing with the improvement of the remaining fatigue life prediction of steel railway bridges, it was possible to perform full-scale fatigue tests on an entire railway bridge. This welded single-span girder bridge has a length of L = 21 m and an open deck, including sleepers and rails. It was taken out of service and should be used to simulate the growth of long fatigue cracks until the final collapse. The paper starts with a detailed presentation of the steel bridge. Afterwards, the individual challenges are shown to get realistic conditions, as expected for this bridge under fatigue loading due to train crossings. Due to the high bending stiffness of the main girders a usual test set-up in the laboratory with one or two single loads was not possible to provide sufficiently high stress-cycles in the flanges of the main girders. Therefore, a hydraulic vibration exciter was used at midspan, to get vibrations of the bridge in the 1st bending eigenfrequency of the main girders. The paper will also address the additional challenges within the fatigue tests and explain how they were overcome. These are: i) provisions to get realistic stress cycles at the bottom flange, only in the tension-tension region, instead of reversed tension-compression cycles, due to the small dead load of the bridge, ii) predictions of the critical constructional details, iii) provisions to enforce long fatigue cracks at only those details within a short test period of 2 weeks. In addition, calculations based on fracture mechanics were done, and the numerical crack growth was compared with the test results at the critical constructional detail. Finally, after man-made elongation of the fatigue cracks, static tests were done to simulate the final collapse of the bridge. The analyses of the stresses within the cross-section of final rupture shows ductile behaviour, instead of brittle failure.