Fuzzy C Means algorithm is popular soft clustering algorithm. It has been applied in many engineering fields. Density weighted FCM is its variant, which can solve FCM’s problem: sensitive to outlier and noise data. However, performances of both algorithms are heavily depend on proper initial cluster centers. This paper proposes a novice algorithm: Improved density weighted FCM based on nearest neighbor pair and its density, simulation results show initial center produced by the algorithm are very close to final cluster center. Thus IDWFCM can convergent very quickly and improve the performance.