The importance of computer simulations in the assessment of nuclear plant safety systems has increased dramatically during the last three decades. The systems of interest include existing or proposed systems that operate, for example, normal operation, in design basis accident conditions, and in severe accident scenario beyond the design basis. The role of computer simulations is especially critical if one is interested in the reliability, robustness, or safety of high consequence systems that cannot be physically tested in a fully representative environment. In the European 7th Framework SARNET project, European Commission (EC) co-funded from 2008 to 2013, the Phébus FPT3 experiment was chosen as a code benchmark exercise to assess the status of the various codes used for severe accident analyses in light water reactors. The aim of the benchmark was to assess the capability of computer codes to model in an integral way the physical processes taking place during a severe accident in a pressurised water reactor (PWR), starting from the initial stages of core degradation, fission product, actinide and structural material release, their transport through the primary circuit up to the behaviour of the released fission products in the containment. The FPT3 benchmark was well supported, with participation from 16 organisations in 11 countries, using 8 different codes. The temperature history of the fuel bundle and the total hydrogen production were well captured. No code was able to reproduce accurately the final bundle state, using as bulk fuel relocation temperature, the temperature of the first significant material relocation observed during the experiment. The total volatile fission product release was well simulated, but the kinetics were generally overestimated. Concerning the modelling of semi-volatile, low-volatile and structural material release, the models need improvement, notably for Mo and Ru for which a substantial difference between bundle and fuel release was experimentally observed, due to retention in the cooler upper part of the bundle. The retention in the primary circuit was not well predicted, this was due mainly the non-prototypic formation of a boron-rich blockage in the rising line of the FPT3 steam generator, simulated in the circuit as a single external cooled U tube. The deposition mechanism and the volatility of some elements (Te, Cs, I) could be better predicted. Containment vessel thermal hydraulics, designed in the experiment to be well-mixed, were well calculated. Concerning the containment aerosol depletion rate, only stand-alone cases (in which the input data were derived from experimental data) provided acceptable results, whilst the integral cases (in which the input data came from circuit calculations) tended to largely overestimate the total aerosol airborne mass entering the containment. The disagreement of the calculated total aerosol airborne mass in the containment vessel with the measured one is due to the combination of a general underestimation of the overall circuit retention and overestimation of fission product and structural material release. Calculation of iodine chemistry in the containment turned out to be a major challenge. Its quality strongly depends on the correct prediction of chemistry speciation in the integral codes. The major difficulties are related to the presence of high fraction of iodine in gaseous form in the primary circuit during the test, which is not correctly reproduced by the codes. This inability of the codes compromised simulation of the observed iodine behaviour in the containment. In the benchmark a significant user effect was detected (different results being obtained by different users of the same code) which had to be taken into account in analysing the results. This article reports the benchmark results comparing the main parameters calculated and observed, summarising the results achieved, and identifying the areas in which understanding needs to be improved. Relevant experimental and theoretical work is under way to resolve the issues raised. © 2016 Elsevier Ltd. All rights reserved.