The random measurement or technological variation of an Affymetrix high‐density oligonucleotide microarray platform is estimated and evaluated against the variation in the gene expression levels of a biological sample. A mixed effects analysis of variance model is used to describe the probe intensity levels. The focus is on the repeatability of the method. Additionally, the minimal number of arrays necessary to detect biological relevant differences in gene expression levels between biological samples is estimated based on the repeatability. The investigated sources of variation for the repeatability are arrays, array production lots, the reverse transcription step, the
in vitrotranscription step and the hybridization and washing step. They may all add to the short‐term random measurement variation. The effect of global linear normalization on the repeatability is determined. It is shown that the in vitrotranscription step contributes to the repeatability with more than 50%. If normalization is used, the microarray platform is indeed suitable for measuring thousands of genes simultaneously and in particular to detect differences in gene expression between biological samples. Based on a goodness‐of‐fit investigation and considering the amount of data involved, the analysis of variance model seems suitable to describe the raw probe intensity data. Copyright © 2005 John Wiley & Sons, Ltd.