Summary.-Evidence for a mutation theory of cancer is presented by reviewing the experimental work on 4-nitroquinoline 1-oxide (4NQO) carcinogenesis. 4NQO almost completely mimics u.v. light and produces 4NQO-purine adducts on DNA. When 4NQO-treated cells are held in liquid medium under appropriate conditions, the 4NQO adducts disappear from DNA, in parallel to decrease of premutational damage in Escherichia coli, or pretransformational damage in cultured mouse cells. Post-treatment with caffeine greatly diminishes the yields by 4NQO of mutants in E. coli, malignant transformants in cultured mouse cells and tumour nodules in the lung of mice. Potentially tumourigenized stem cells in the lung remain sensitive to selective killing by caffeine for at least 5 days after 4NQO treatment, in spite of their DNA being apparently replicated, an indication that carcinogen-damaged DNA in the stem cell can be transmitted to its successive daughter stem cells for many generations. This peculiar characteristic is discussed as a possible lead to the crux of the mutation theory of cancer in vivo, and a model for carcinogenesis is proposed.