The deoxycytidine analog 5-aza-2'-deoxycytidine (5-azadCyd) has been widely used as a DNA methylation inhibitor to experimentally induce gene expression and cellular differentiation. Prior to the availability of mutant mice with altered DNA methyltransferase levels, treatment of cells with drugs has been the only means to experimentally manipulate the level of genomic DNA methylation in mammalian cells. Substitution of DNA with 5-azadCyd leads to covalent trapping of the enzyme, thereby depleting the cells of enzyme activity and resulting in DNA demethylation. 5-AzadCyd or 5-azacytidine treatment causes multiple changes in treated cells, including activation of silent genes, decondensation of chromatin, and induction of cellular differentiation, all of which are believed to be consequences of drug-induced demethylation. 5-AzadCyd is highly toxic in cultured cells and animals and is utilized as a potent antitumor agent for treatment of certain human cancers. It has been postulated that the toxicity of the drug in mammalian cells is also due to its inhibition of DNA methylation. The chemistry of the methylation reaction is consistent, however, with an alternative mechanism: the cytotoxic effect of 5-azadCyd may be directly mediated through the covalent binding of DNA methyltransferase to 5-azadCyd-substituted DNA. We have tested this possibility by using embryonic stem cells and mice with reduced levels of DNA methyltransferase due to a targeted mutation of the gene. When exposed to 5-azadCyd mutant embryonic stem cells or embryos were significantly more resistant to the toxic effects of the drug than wild-type cells and embryos, respectively. These results strongly suggest that the cellular DNA methyltransferase itself, rather than the secondary demethylation of genomic DNA, is the primary mediator of 5-azadCyd cytotoxicity. In light of our results, some conclusions from previous studies using 5-azadCyd in order to experimentally manipulate cellular methylation levels may have to be reassessed. Also, our data make clear predictions for cancer treatment: tumor cells with elevated DNA methyltransferase levels would be expected to be susceptible to treatment with 5-azadCyd, whereas tumors with reduced levels of the enzyme would be resistant.