Accuracy assessment is an integral part of the production of land cover/land use maps. The process requires the availability of a good-quality validation dataset for the qualitative and quantitative evaluation of generated products. This paper describes the development of the validation dataset that was used for the accuracy assessment of the Land Cover Map of Europe 2017 in the context of the Sentinel-2 Global Land Cover project. Sample selection was based on a two-step stratified random sampling process. In the first step, validation sites (Sentinel-2 tiles) were selected randomly and in proportion to the area covered by each country. In the second step, validation sites were stratified with the CORINE Land Cover dataset, which enabled the proportional selection of land cover classes. The selected samples were visually inspected by experts who categorised them into 13 classes. This resulted in a large set of 52,024 samples, spread over Europe. The final dataset can be used to validate European land cover products at continental scale, and may also be included in larger (e.g. global) datasets, or for country-based studies.