Microarray technologies are generating new data at a breathtaking speed. Because of the sheer size of these data and the technical complexity involved in microarray chip designs, it is a challenge to assess the reliability and the accuracy of these data. In our research using Affymetrix microarrays, many genes of interest are expressed at low levels and are only slightly up- or down-regulated under various biological conditions. By using the software offered by Affymetrix Inc., these genes are mostly called “absent”. However, in the original raw data on the chip images, we have found consistent and reproducible patterns regarding those lowly expressed genes. Based on this finding we designed and tested a new algorithm for expression profile evaluation. When applied to our microarray data set, the new method seems to be able to identify genes that are simply called “absent” by Affymetrix's method and actually have reproducible changes in gene expression in duplicate experiments. Furthermore, compared with Affymetrix's algorithm, the new algorithm reports far fewer genes as differentially expressed when comparing two samples that were prepared under identical biological conditions. Thus, the new method is believed to have less false positives and more true positives in recognising differentially expressed genes, a highly desired property for the microarray data analysis.