Abstract: cDNA microarray expression data is widely used to help biomedical research. There are so many challenges when the computer and pattern recognition methods are used to process and analysis the data. In this paper, a novel hybrid cross-phenotype normalization method is proposed which deals with supervised multi-class multi-sample cDNA expression data set based on invariantly expressed genes. The algorithm can be directly used as a feature selection method for gene classifier. The result is satisfactory.