Normalization of cDNA and oligonucleotide microarray data has become a standard procedure to offset non-biological differences between two samples for accurate identification of differentially expressed genes. Although there are many normalization techniques available, their ability to accurately remove systematic variation has not been sufficiently evaluated. In this study, we performed experimental validation of various normalization methods in order to assess their ability to accurately offset non-biological differences (systematic variation). The limitations of many existing normalization methods become apparent when there are unbalanced shifts in transcript levels. To overcome this limitation, we have proposed a novel normalization method that uses a matching algorithm for the distribution peaks of the expression log ratio. The robustness and effectiveness of this method was evaluated using both experimental and simulated data.