Large numbers of incident related databases have been established in the last three decades. The majority of attempts to explore these data marts were trials to identify patterns via first glance into the datasets. This study investigated a subset of incidents from fixed facilities in Harris County, TX, extracted from the National Response Center database. By classifying the information into groups and using data mining techniques, interesting patterns of incidents according to characteristics such as type of equipment involved, type of chemical released and causes involved were revealed and further these were used to modify the annual failure probabilities of equipments. (c) 2005 Elsevier B.V. All rights reserved.