Abstract: Because gene expression data is high dimensions and small samples，especially the less priori knowledge，a two-way clustering algorithm based on the representative entropy is proposed，which is combined with the advantages of Self Organizing feature Map（SOM） neural network.First，the clustering of genes is realized through the SOM network，and characteristic genes are selected according to the fluctuation coefficient.Then the quality of gene clustering is decided by the value of representative entropy.Finally，Self Organizing Feature Map algorithm is employed to classification of samples.This process is applied to two published data sets of gene expression.The experiment results show that the algorithm can reduce the feature space dimensions and improve the accuracy of clustering.