A high percentage of false positives remains a problem in current network security detection systems. With the growing reliance of industry on computer networks, and the growing variety of attacks that can be directed towards a computer network, it is clear that detection systems must be improved in order to tackle this growing problem. To help minimise the problem of false positives, this paper describes a method and apparatus for security alert analysis that is based on two technologies: (i) event correlation and (ii) a truth maintenance system. This work was undertaken in the context of practical network security management in a large outsourced management service provider in the Asia-Pacific region.