Labeling the connected components in the feature space is an important step in grid based clustering algorithms in data mining. Although Connected Components Labeling Algorithms have been highly improved in image processing domain, there is little progress in grid based clustering in data mining domain. Two problems exist in transplanting these algorithms from image processing to data mining. One is how to process multi-dimensional dataset. The other is how to reduce the cost of auxiliary space. This paper describes an optimal two-scan Connected Components Labeling algorithm based that in image processing domain. It does not need auxiliary space, and easy to be extended to multi-dimension data set.