Classification accuracy is tested using six functional classes defined by the MIPS Yeast Genome Database (see Table 1). The MYGD class definitions come from biochemical and genetic studies of gene funtion, while the microarry expression data measures mRNA levels of genes. Many classes in MYGD, especially structural classes such as protein kinases, will be unlearnable from expression data by any classifier. The first five classes in Table 1 are selected because they represent categories of genes that are expected, on biological grounds, to exhibit similar expression profiles. Furthermore, Eisen et al. showed that these classes cluster well using hierarchical clustering based upon the normalized dot product [Eisen et al., 1998]. The sixth class, the helix-turn-helix proteins, is included as a control group. Since there is no reason to believe that the members of this class are similarly regulated, we do not expect any classifier to learn to recognize members of this class based upon mRNA expression measurements.