Two extensive measurement campaigns have been carried out to validate the TWINS prediction program for rolling noise, as described in part 1 of this paper. This second part presents the experimental results of vibration and noise during train pass-bys and compares them with predictions from the TWINS program. The model is found to give quite reliable results provided that care is taken with the input data, especially that concerning the roughness and the attenuation of vibration along the rail. It is found that the model consistently predicts overall levels about 2 dB(A) higher than the measured values. This effect can be attributed to simplifications in the model for the propagation of sound from source to receiver. On the other hand spectral results reveal a large spread in each one-third octave band (standard deviations of up to 5 dB). This can be identified with a similar level of uncertainty concerning the actual roughness acting at the contact patch, due to uncertainties in the position of the contact patch and variations between the roughness of nominally similar wheels and rail sections.