As companies move progressively toward quantifying the risks of releases of hazardous materials, there becomes a greater need for developing the data necessary to populate the risk analysis. Sophisticated mathematical models have been developed to predict the consequences of a hazardous material release. But the effort devoted to the frequency side of the "risk equation" has been very disorganized by comparison, with inconsistent or non-existent definitions of "failure", mixing of incompatible data, application of data from one industry to a completely different industry, and a host of other problems. Nonetheless, through judicious assembly and analysis of a variety of data sources, a useful failure rate database can be developed. Many seminal sources of data are described, with an emphasis on loss of containment failure rates. Pitfalls in interpreting failure rate data are also illustrated. (c) 2005 Elsevier B.V. All rights reserved.