Extensive studies of the H-mode density limit (DL) in JET gas-puffed discharges have been performed in the past four years targeting at an improved database for extrapolation to ITER. This paper reviews the arguments for the particular DL definition (pedestal density at the H–L boundary), the logic underlying the choice of parameters under focus (toroidal field, major radius, triangularity, safety factor) as well as some improvement in the interpretation of typical JET density ramp-up signatures that led to a critical review of the existing data. An empirical scaling is derived and compared with existing empirical and model based scalings. ASDEX Upgrade data are included in this analysis to provide information on the size dependence. The main results are: earlier findings on the Bt, R and q95-dependences are confirmed. The triangularity dependence, if any, is weak. The SOL-based BLS (Borrass, Lingertat, Schneider) scaling and the empirical scaling are virtually indistinguishable. The Greenwald scaling provides values in the right absolute range, but the overall fit is comparatively poor. The proposed scaling predicts ITER critical densities considerably below the reference value. Fuelling methods other than gas-puffing are outside the scope of this paper, but pellet fuelling, envisaged for ITER as an option to alleviate the situation, is discussed.