SAGE is a powerful tool to analysis whole-genome expression profiles. For improving the accuracy and efficiency of pattern recognition and clustering analysis, SAGE data is needed to be reducing dimensions due to its large quantities and high dimensions. A Poisson-Model based kernel (PMK) was proposed based on the Poisson distribution of the SAGE data. Kernel Principle Component Analysis (KPCA) with PMK was used in reducing dimensions analysis of mouse retinal SAGE data. The experimental results show that it can eliminate data redundancy effectively and reduce dimensions.