Clustering over gene expression is now a popular computational analysis in biology. In general, the amount of expression can be measured by high-throughput techniques over thousands of genes simultaneously. The expression dataset can be a large table (or matrix) with numerical values, each being specified by one gene and one sample, and needs computational methods to be analyzed. This review starts with surveying techniques of clustering genes by expression, classifying them into three types: hierarchical, partitional, and subspace clustering. Major methods of hierarchical and partitional clustering as well as a variety of algorithms for subspace clustering are extensively reviewed. Techniques for clustering over expression, however, are now well matured and their performance is limited due to the inevitable noisiness of the high-throughput nature of expression data. We then extend the scope of this review further to clustering genes with recently emerging data, gene networks, and show graph partitioning approaches, such as spectral methods, for clustering genes by a network. Furthermore, advanced approaches of gene clustering now combine gene networks with expression. This setting corresponds to so-called semi-supervised clustering in machine learning, and approaches under this problem setting will be widely reviewed, classifying those approaches into three types. (c) 2011 John Wiley & Sons, Inc. WIREs Data Mining Knowl Discov 2011 1 496-511 DOI: 10.1002/widm.41