This paper aims at proposing an approach that can detect feature interactions through patterns analysis. The basic idea is to extract the common conflict patterns from the known feature interactions and reuse them to detect the unknown ones. With this approach, the conditions for conflicts are described using a set of predicate formulae and the system model is specified in Java language. With the aid of an external tool, all the execution scenarios can be systematically explored. During the execution of the system model, all the feature behaviors will be collected and analyzed. Once a conflict condition is satisfied, a conflict report is produced. The approach has been applied to an E-mail system. The experimental results show that the approach can effectively detect feature interactions, both the known and the unknown, after handling more than 1 million execution scenarios.