Machinery maintenance activities should be based on statistics and knowledge of past machinery failures, but it is often difficult to keep gathering those data in engineers' daily works. This paper describes an approach to acquire those data and knowledge from machinery failure reports, which are written as free texts, by applying text mining techniques. The method consists of three stages. The first state is event extraction. The term event represents a status of an object;“damage of a piston ring”is an example of it. Each machinery failure reports is converted into a set of events in this stage. The second stage is to find relations between the events. To find taxonomic relations, we defined a lexical distance and direction measure to evaluate similarities among events. To find non-taxonomical relations, an association rule mining method is used. The third stage is evaluating extracted relations. Authors tested the proposing knowledge acquisition process with 136 machinery failure reports and expert engineers validated more than 80% relations are correct knowledge.