Expression profile data for all the genes in a genome can be obtained by transcriptome experiments such as GeneChip and cDNA array systems. Classification of genes in high resolution based on expression profiles may become a key to understand the integrated system in a cell. To attain this purpose, we conduct butch-learning self-organizing map (BL-SOM) for classifying genes based on expression profiles. In the present study, we consider two-types of SOMs based on configuration of units. One of them has planar representation space called "Plain SOM" and the other has spherical space called "Spherical SOM". In general, the profile vectors of gene expression are normalized to unity in length in order to focus not on the absolute quantity of expression but the similarity of the direction. In the present study, we propose a measure for the suitability of SOM to a given data set and confirm that the Spherical SOM is actually suitable for the data on gene expression. Then, we examine biological meanings of gene classification of Bacillus subtilis.