Experiments using microarrays abound in genomic research, yet one factor remains in question. Without replication, how much stock can we put into the findings of microarray experiments? In addition, there is a growing desire to integrate microarray data with other molecular databases. To accomplish this in a scientifically acceptable manner, we must be able to measure the validity and quality of microarray data. Otherwise, it would be the weakest link in any integration process. Validating and evaluating the quality of data requires the ability to determine the reproducibility of results. Data obtained from a microarray experiment designed as a feasibility test provided a unique opportunity to partition and quantify several sources of variation that are likely to be present in most microarray experiments. We use this opportunity to discuss the origins of variability observed in microarray experiments and provide some suggestions for how to minimize or avoid them when designing an experiment.