Concept extraction work, promises to improve the performance of the term-based text mining which has high complexity. The first phase of the concept extraction is to detect the terms have notable frequency to represent the documents. With grouping these terms an important function will be implemented on the way conception. Transition from terms to concepts; by clustering the terms according to similarities between terms, and then by labeling these clusters with an expert. The parameters of clustering algorithm and the quality of the data set will affect the success of this process. In this study, the three methods for term similarity are examined and the the most successful one is tried to find. Study is performed on Turkish language.