Microarray technology is a powerful tool for generating expression data on a large number of genes simultaneously. However, as for any assay, it must be reproducible to give confidence in the results. Using a classical statistical method--the factorial design of experiments--we have assessed the effects of different experimental factors in our system. Significant effects on signal were seen when the standard components were substituted with a different enzyme, fluorescent label, or RNA purification method. This has led to the implementation of an improved procedure that maximizes signal without affecting the variability of the system, thus increasing the signal-to-noise ratio. In addition, we were able to quantify the variability between microarrays and replicates within microarrays.