Abstract: In order to study the relation between tumor and gene, the authors analyze the small round blue cell tumor(SRBCT) gene expression data and establishes multi-models prediction model(MMPM) for classifying and predicting four families of SRBCT. MMPM creates four classifiers based on Neural Networks to classify four families and codes results of four classifiers to final prediction consequence for each sample. Research indicates that to divide multi-class classification into several binary classifications is an effective method solving multi-class classification and that MMPM based on this method can learn knowledge from gene expression data and can exactly classify and predict all samples.