Traditional fuzzy clustering based segmentation algorithms are sensitive to noise. Therefore, an improved fuzzy C-means clustering(FCM) algorithm based on neighborhood similarity is proposed. Firstly, similarities between the center pixel and its neighbour pixels are defined from the spectral and membership characteristics, respectively. Secondly, the neighbor constraint item is defined combining the two similarities and the distance from each neighborhood pixel to the cluster centers. Then the objective function of the proposed mixed neighborhood constraints based fuzzy C-means(MNCFCM) algorithm is defined by adding the neighbor constraint item in order to keep balance between image smoothing and details preserving during segmentation. Finally, through qualitative and quantitative evaluation of the segmentation results of the composite image and the real remote sensing image, it is verified that the algorithm is robust to noise and can preserve image details at the same time, which can obtain highly accurate segmentation results.