DNA microarray users face many challenges to obtain accurate results, including complex technical errors, natural variability of biological systems, imperfect reproducibility of reference standards, and difficulties in acquisition and processing of large amounts of data. Therefore, investigators should be aware of potential sources of variability and account for them in the experimental design and execution. This work reports our experience in identifying factors that alter the reliability of the results and in diminishing effects of these factors. We have studied the hybridization reproducibility in cDNA microarray chips, both as absolute values and expression ratios, and the nature and impact of several technical, acquisition, and processing errors. A new experimental strategy is proposed and mathematical algorithms developed that efficiently correct the errors and thereby increase the information obtainable through microarray studies. These algorithms reduced the variability not associated with biological system to less than a quarter of its initial value and have substantially enhanced reliability in experiments on brain and cultured neuroblastoma cells.