Block propagation models have been used for years for rockfall hazard assessment. However, the settings of model parameters that guarantee the predictive capabilities of the simulations for a given study site remains a key issue. This research aims at investigating the predictive capabilities of block propagation models after a preliminary calibration phase. It is focused on models integrating the shape of blocks since, despite their sound physical bases, they remain less used than lumped-mass approaches due to their more recent popularisation. Benefiting from both a recently built model integrating block shape usable in 2D and 3D and from recent experimental results at the slope scale, we first performed an expert-based calibration based on the use of the 2D model and, second, evaluated the predictive capabilities of the calibrated model in 2D and in 3D using the remaining part of the experimental results. The calibrated model simulations predict the main characteristics of the propagation. Similar level of adequacy between simulations and experimental results in the calibration and validation phases emphasizes genericity of the model settings which guarantees practical applicability : after a calibration phase on sufficient amount of soil types, the model may be used in a predictive manner. The adequacy between 2D and 3D simulations also favors applicability of the model since easier and faster calibrations based on 2D simulations only can be envisaged. As classically observed for block propagation models, the model is not sufficient to predict the details of the velocity and stopping points but provides accurate prediction of the global ranges of these quantities, in particular of the extreme values. To lift these limitations in terms of predictive capabilities, more advanced calibration procedures based on optimization techniques can constitute a promising perspective.