The calibration of the electrical performance of seven photovoltaic (PV) modules was compared between four reference laboratories on three continents. The devices included two samples in standard and two in high-efficiency crystalline silicon technology, two CI(G)S and one CdTe module. The reference value for each PV module parameter was calculated from the average of the results of all four laboratories, weighted by the respective measurement uncertainties. All single results were then analysed with respect to this reference value using the E-n number approach. For the four modules in crystalline silicon technology, the results agreed in general within +/- 0.5%, with all values within +/- 1% and all E-n numbers well within [-1,1], indicating further scope for reducing quoted measurement uncertainty. Regarding the three thin-film modules, deviations were on average roughly twice as large, i.e. in general from +/- 1% to +/- 2%. A number of inconsistent results were observable, although within the 5% that can be statistically expected on the basis of the En number approach. Most inconsistencies can be traced to the preconditioning procedure of one participant, although contribution of other factors cannot be ruled out. After removing these obvious inconsistent results, only two real outliers remained, representing less than 2% of the total number of measurands. The results presented show improved agreement for the calibration of PV modules with respect to previous international exercises. For thin-film PV modules, the preconditioning of the devices prior to calibration measurements is the most critical factor for obtaining consistent results, while the measurement processes seem consistent and repeatable. (C) 2017 The Authors and National Renewable Energy Laboratory. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.orgilicensesiby/4.0/).