MARS is the system established and maintained by the European Commission in order to collect information related to major industrial accidents in EU Member States in the context of the Seveso II Directive. One of the main purposes of the MARS database is to provide information for learning from the accidents to avoid similar events. Probably, the most important issue for the learning is the determination of the causes, particularly the underlying causes, of the accidents. One objective was to find possible patterns of underlying causes per industry type and per country. Another objective was to determine the occurrence of weaknesses in safety management systems and in safety culture as underlying causes. A further objective was to determine the level of learning from the accidents, as it appears from the reports, per industry type and per country. A sequential method, presented by us in a previous paper in this publication, was used to make it possible to go beyond the causes given in the original reports and to find more underlying causes. To determine the level of learning from the accidents, using the actions/lessons learned given in the reports, a classification method was developed. This method establishes the level of learning of the lessons learned from each case description, essentially from the organisational point of view. This paper presents the results of an analysis regarding underlying causes of all the accidents of the MARS database reported up to mid 2007. The results are expressed per industry type and per country. The main results are that as much as three times as many underlying causes can be found when applying the method developed compared with what is given in the original reports. The most important underlying causes are found in weaknesses in process analysis (risk assessment) and in procedures, regardless of industry type. Weaknesses in safety management systems and in safety culture contribute as underlying causes in a very high percentage of the accidents. The quality of reporting, measured in terms of analysis of underlying causes, vary considerably between various countries. The level of learning, as determined from the information in the reports, is found to be in general rather low, especially from some of the countries. This study has given rise to ideas of improvement of the MARS system. It has also raised many questions, some of which would be suitable for further research. (C) 2009 Elsevier Ltd. All rights reserved.