Evolutionary studies have suggested that mutation rates vary significantly at different positions in the eukaryotic genome. The mechanism that is responsible for this context-dependence of mutation rates is not understood. We demonstrate experimentally that frameshift mutation rates in yeast microsatellites depend on the genomic context and that this variation primarily reflects the context-dependence of the efficiency of DNA mismatch repair. We measured the stability of a 16.5-repeat polyGT tract by using a reporter gene (URA3-GT) in which the microsatellite was inserted in-frame into the yeast URA3 gene. We constructed 10 isogenic yeast strains with the reporter gene at different locations in the genome. Rates of frameshift mutations that abolished the correct reading frame of this gene were determined by fluctuation analysis. A 16-fold difference was found among these strains. We made mismatch-repair-deficient (msh2) derivatives of six of the strains. Mutation rates were elevated for all of these strains, but the differences in rates among the strains were substantially reduced. The simplest interpretation of this result is that the efficiency of DNA mismatch repair varies in different regions of the genome, perhaps reflecting some aspect of chromosome structure.