This paper introduces a method to cluster graphs of semantically related terms from texts using PageRank calculations for use in the field of text mining, e.g. to automatically discover different topics in a text corpus. It is evaluated by providing empirical results of tests by applying this method on real text corpora. It is shown that this application of the PageRank formula realizes suitable clustering such that the mean similarity between the terms in the clusters reaches a high level. A special state transition in the mean term similarity is discussed when analysing texts with stopwords.