There are many undesirable systematic variations in general microarray experiments. Normalization is the process of removing such variations which affect the measured gene expression levels. The aim of an effective normalization method for cDNA microarray data is to remove noise while retaining the signal in the data. A composite three-stage normalization method was proposed in this paper. The first stage and second stage normalization method will be done within-slide, and the third stage is between-slide. The first stage is print-tip lowess normalization, it provides a well-tested general purpose normalization method which has given good results on a wide range of arrays. In the second stage, we use shift-log normalization method to treat the data that we have gotten from the first stage. The dye bias can be balanced in this stage. In the third stage, we will make the M value normalized by using the median absolute deviance (MAD) value. Although the three normalization methods are well-known, the combination of them, so called composite method turns out to be more effective than any single normalization method. We attempt to make the M vs A plot to be a horizontal band centered around 0. The composite three-stage normalization can carry out our intention.