Forecasting the behaviour of a flammable or toxic cloud is a critical challenge in quantitative risk analysis. Recent literature shows that empirical and integral models are unable to model complex dispersion scenarios, like those occurring in semi-confined spaces or with the presence of physical barriers. Although CFD simulators are promising tools in this regard, they still need to be fully validated with comprehensive datasets coming from experimental campaigns designed ad-hoc. In this paper, we present an experimental campaign carried out by a joint venture between University of Sao Paulo and Universitat Politecnica de Catalunya to investigate CFD tools performance when used to analyse clouds dispersion. The experiments consisted on propane cloud dispersion field tests (unobstructed and with the presence of a fence obstructing the flow) of releases up to 0.5 kg s(-1) of 40s of duration in a discharge area of 700 m(2). We provide a full 1-s averaged propane concentration evolution dataset of two experiments, extracted from 29 points located at different positions within the cloud, with which we have tested FLACS (R) CFD-software performance. FLACS reproduces successfully the presence of complex geometry, showing realistic concentration decreases due to cloud dispersion obstruction by the existence of a fence. However, simulated clouds have not represented the whole complex accumulation dynamics due to wind variation. (C) 2015 Elsevier Ltd. All rights reserved.