Boiling flows are frequently found in industry and engineering due to the large amount of heat that can be transferred within such flows with minimum temperature differences. In the nuclear industry, boiling affects in different ways the operation of almost all water-cooled nuclear reactors. Recently, the use of computational fluid dynamic (CFD) approaches to predict boiling flows is increasing and, in the nuclear area, CFD is being developed to solve thermal hydraulic safety issues such as establishing the critical heat flux, which is perhaps the major threat to the integrity of nuclear fuel rods. In this paper, the accuracy of an Eulerian-Eulerian, two-fluid CFD model is evaluated over a large database of subcooled boiling flows, avoiding the rather popular case-by-case tuning of descriptive models to a limited number of experiments. The model includes a Reynolds stress turbulence model, the method of moments-based S-gamma population balance approach and a boiling model derived using the heat flux partitioning approach. The database covers a large range of conditions in subcooled boiling flows of water and refrigerants in vertical pipes and annular channels. Overall, a satisfactory predictive accuracy is achieved for some quantities of interest, such as the void fraction and the turbulence and liquid temperature fields, but results are less satisfactory in other areas, more specifically for the average bubble diameter and the mean velocity profiles close to the wall in annular channels. Agreement may be improved with advances in the treatment of large bubbles and bubble break-up and coalescence, as well as in improved modelling of the boiling region close to the wall, and more specifically the bubble departure diameter, the wall treatment and the contribution of bubbles to turbulence. (C) 2016 Elsevier Ltd. All rights reserved.