Modelling bubble columns operating in the churn turbulent flow regime is a considerable challenge due to their complex and transient behaviour. A key modelling decision is how best to model the bubble size distribution, with a range of approaches existing from the relatively simple option of using a single, representative bubble size to more complex population balance approaches. In this comparative study, we have examined various approaches (i.e. constant and variable single bubble size models, dual bubble size models, and both homogeneous and inhomogeneous MUSIG models). Model predictions were compared with experimental data from three different configurations of a pilot-scale column (0.39 m in diameter and 2 m in height) operated at industrially relevant superficial gas velocities of 0.16 and 0.25 m s(-1). The single bubble size models gave reasonable agreement with experimental data for the 'symmetrical' configuration, with the variable bubble size model having the major advantage of providing acceptable predictions of the mean bubble size (without this having to be provided as a model input). Implementation of multiple bubble size models did not lead to any substantial improvement in the modelling fit with the experimental data, leading to the inescapable conclusion that the standard bubble breakup and coalescence kernels proposed in the literature do not adequately describe bubble transformations taking place under heterogeneous operating conditions. (C) 2018 Elsevier Ltd. All rights reserved.