Normalization methods for the pre-process of gene chip are an important step.By studying normalization methods for gene chip,an improved algorithm is proposed in this paper.The improved algorithm has nearly effect to the robust Lowess normalization method, however our method spend less running time.The method is smoothed to the data points by locally weighted linear regression.Then kernel estimation is used for estimation of residue from smoothing estimation.Finally,scaling each data point on the plaid. Experimental results show that the algorithm is effective.