Abstract: In order to deal with the problem that the traditional clustering methods were depended on clustering features and were difficult to find the features which can represent completely clusters, remote sensing image segmentation method combining multivariate information clustering with spatial constraints is proposed by considering the spatial position relation of remote sensing data. Given a cluster, several data points in the cluster are selected to form multi-tuple and all possible multi-tuples are employed as features to represent the cluster; And then mutual information of grayscales of pixels within each multi-tuple is calculated to characterize the similarity of the cluster; the dissimilarity between two clusters is defined by the mutual information of groups of pixels including multi-tuple of one cluster and pixels from another cluster. The objective function is formulated by balancing the above similarity and disdimularity. By maximizing the objective function, remote sensing image segmentation can be obtained. Comparing the segmentation results of simulated and real remote sensing image from qualitative and quantitative analyses indicate that the proposed algorithm can avoid the definition of clustering features, so it fundamentally eliminates its influence on image segmentation and considers adequately the spatial position relation.