Nowadays, optical network turns to be more complex. Once there occurs a failure, there will result in multi-alarm-events. On the other hand, an alarm might result from different faults. This paper describes the alarm correlation in optical networks based on data mining. We mine association rules for dynamic network resource and service by fully utilizing knowledge base formed before, which enables the framework can be easily updated and new discovery methods can be readily incorporated within it. Fuzzy logic is also used to convert numerical alarm attributes into linguistic terms. Once there are new unknown alarms occur in network, we use fuzzy reasoning based on fuzzy association rules (FARs) for fault diagnosis. It will shorten the recovery time and improve the performance of optical network fault management. Experiments are carried out to validate the accuracy &efficiency of our research.