Chemical mutagenesis has been the workhorse of traditional genetics, but it has not been possible to determine underlying rates or distributions of mutations from phenotypic screens. However, reverse-genetic screens can be used to provide an unbiased ascertainment of mutation statistics. Here we report a comprehensive analysis of approximately 1900 ethyl methanesulfonate (EMS)-induced mutations in 192 Arabidopsis thaliana target genes from a large-scale TILLING reverse-genetic project, about two orders of magnitude larger than previous such efforts. From this large data set, we are able to draw strong inferences about the occurrence and randomness of chemically induced mutations. We provide evidence that we have detected the large majority of mutations in the regions screened and confirm the robustness of the high-throughput TILLING method; therefore, any deviations from randomness can be attributed to selectional or mutational biases. Overall, we detect twice as many heterozygotes as homozygotes, as expected; however, for mutations that are predicted to truncate an encoded protein, we detect a ratio of 3.6:1, indicating selection against homozygous deleterious mutations. As expected for alkylation of guanine by EMS, >99% of mutations are G/C-to-A/T transitions. A nearest-neighbor bias around the mutated base pair suggests that mismatch repair counteracts alkylation damage.