In Gasoline Direct Injection (GDI) engines, the secondary break-up plays a significant role in air–fuel mixing. In fact, spray granulometry affects evaporation rate, liquid penetration and plume morphology. Operating pressures and temperatures of both liquid and gaseous phases strongly influence the droplet disruption mechanism in the combustion chamber. In 3D Computational Fluid Dynamics (CFD), several models can be adopted to simulate the secondary break-up process, among which the Reitz-Diwakar and the Kelvin-Helmholtz Rayleigh-Taylor (KHRT) are the most diffused ones in the engine community. However, application of such models in their original versions is limited to a reduced range of injection parameters and ambient conditions. As a matter of fact, large variations of them usually require ad-hoc calibrations of the model constants. To improve the predictive capabilities and reduce the need of case-by-case tuning, an alternative secondary break-up model is proposed in the present paper. It is based on the Reitz-Diwakar one but, compared to the latter, a zonalization of the break-up regimes is proposed. Specifically, it is assumed that only Stripping break-up can occur near the nozzle, while Bag break-up only takes place sufficiently far from it. Moreover, model parameters are now treated as functions of the operating conditions. In particular, the impact of the ambient density on the model parameters is analysed in the present work. The proposed model is calibrated via constant volume vessel simulations on a single-hole GDI research injector at vacuum-to-pressurized conditions (namely at 0.4, 1.0, and 3.0 bar(a) of back pressure), on equal temperature. Model parameters are found to be linear functions of the ambient density. Thereafter, model validation is carried out on two different GDI injectors. The first is again a single-hole (remarkably different compared to the previous one), while the second is a 5-hole prototype. Numerical results provided by the proposed model show a satisfactory agreement compared to the experiments in terms of liquid penetration, Phase Doppler Anemometry (PDA) data and imaging, without any dedicated tuning. Conversely, the Reitz-Diwakar and KHRT models, applied to simulate (with default calibration constants) the same injectors, provide results which remarkably deviate from the experiments.