Abstract: Aim at these problems of most of remote sensing image data don’t submit to gauss distribution, as well as remote sensing image classification exists the nonlinear, fuzzy and few labeled data, a semi-supervised kernel-based fuzzy C-means (SSKFCM) algorithm is proposed for classification of multispectral remote sensing image. First, the SSKFCM algorithm is presented by introducing simultaneously semi-supervised learning technique and kernel method into conventional fuzzy C-means (FCM) algorithm. Then, the experiments of the SSKFCM algorithm and a few conventional classification methods are implemented to test the properties of classification results for Beijing-1 micro-satellite’s multispectral image. Finally, the classification performance is estimated by corresponding indexes. The results show that the SSKFCM algorithm improves significantly classification accuracy compared with conventional classifiers (K-means algorithm and maximum likelihood algorithm). Also, it outperforms the FCM algorithm, the KFCM algorithm and the SSFCM algorithm.