SOA has become more and more popular, but fault tolerance is not supported in most SOA-based applications yet. Although fault tolerance is a grand challenge for enterprise computing, we can partially resolve this problem by focusing on its some aspect. This paper focuses on fault detection and puts forward a queueing-theory-based fault detection mechanism to detect the services that fail to satisfy performance requirements. This paper also gives a reference service model and reference architecture of fault-tolerance control center of Enterprise Services Bus for SOA- based applications.