Expression profiling using DNA arrays is often believed to have appeared during the second half of the 1990s, and to be based exclusively on nonisotopic methods. In fact, the first article describing the application of cDNA arrays to expression analysis was published in 1992, relied on radioactive labeling, and was a new development of "high-density" membranes used until then essentially for efficient screening of libraries. Several papers described the use of this technology for simultaneous expression measurement of thousands of genes at the time when the first glass microarrays were published. Simultaneously, oligonucleotide chips, originally developed for resequencing and mutation detection applications, were shown to be capable of expression measurement as well. The three approaches have developed over the years and still coexist, as each of them has specific advantages (and drawbacks); the major issues have become those of data quality, data analysis and storage (ideally in a common public database). Meanwhile, the technology continues to evolve. The most obvious trend is a shift towards using arrays of relatively long oligonucleotides that combine most of the advantages of very long (cDNA) and very short (25-mer) DNA segments. The search for better detection methods, ideally without labeling of the sample, is continuing, although it seems difficult to reach the required sensitivity. New materials for microarray manufacture and new implementations of existing methods have appeared. In addition, the field is progressively becoming segmented into high gene number, low volume (research) applications on the one hand, and low gene number, high throughput (diagnostic) uses on the other.