Several collisional–radiative (CR) models (Anderson et al 2000 Plasma Phys. Control. Fusion 42 781–806, Hutchinson 2002 Plasma Phys. Control. Fusion 44 71–82, Marchuk et al 2008 Rev. Sci. Instrum. 79 10F532) have been developed to calculate the attenuation and the population of excited states of hydrogen or deuterium beams injected into tokamak plasmas. The datasets generated by these CR models are needed for the modelling of beam ion deposition and (excited) beam densities in current experiments, and the reliability of these data will be crucial to obtain helium ash densities on ITER combining charge exchange and beam emission spectroscopy. Good agreement between the different CR models for the neutral beam (NB) is found, if corrections to the fundamental cross sections are taken into account. First the Ha and Hß beam emission spectra from JET are compared with the expected intensities. Second, the line ratios within the Stark multiplet are compared with the predictions of a sublevel resolved model. The measured intensity of the full multiplet is ˜30% lower than expected on the basis of beam attenuation codes and the updated beam emission rates, but apart from the atomic data this could also be due to the characterization of the NB path and line of sight integration and the absolute calibration of the optics. The modelled n = 3 to n = 4 population agrees very well with the ratio of the measured Ha to Hß beam emission intensities.