The recent advent of oligonucleotide microarray technology has made genomic-scale exploration of gene expression possible. Large volume of data has been generated rapidly and needs novel statistical and machine-learning methods for analysis. We propose a model-based expression calculation method to reduce the noise at low-level analysis; the validity of the model is assessed in several aspects and the standard errors attached with the expression values are applied to confidence interval calculation and hierarchical cluster re-sampling. A software package “dChip” is written to facilitate scientists to rapidly apply the methods and obtain biological discovery.