With the rapid expansion of information, scale and dimensionality of data are constantly increasing. Traditional clustering methods are difficult to adapt to this trend. Especially, given the fast development of mobile computing platforms, its properties limit the scale of memory that algorithms can use, so many algorithms cannot run on such platforms without making improvements. This paper proposes a clustering method based on nearest neighbor representation. This method uses the idea of nearest neighbors to construct the new representation. This new representation is compressible, thus effectively reducing the storage cost required for clustering. An algorithm called Bit k-means in implemented to perform clustering directly on the compressed nearest neighbors representation. Experimental results show that the new method achieves higher accuracy and substantially reduces the storage cost.