Event extraction is a major task of Automatic Content Extraction (ACE) program. This paper focuses on the sub-task of event extraction, event argument identification, and proposes a novel method for Chinese event argument identification. The method involves two steps: (1) weighting features by the ReliefF algorithm for considering the particular contributions of different features on clustering analysis, and (2) employing a semi-supervised clustering algorithm, Constrained-KMeans, to group event arguments. Compared with normal Constrained-KMeans algorithm, feature weighting obviously improves the F-Measure of identification. The comprehensive experimental results also demonstrate the outstanding performance of the new method.