Understanding the genomic impact of carcinogens is fundamental to cancer biology and prevention. However, recent coordinated efforts to detect such fingerprints have been largely unsuccessful, challenging the paradigm that carcinogens induce identifiable mutational signatures. Here we introduce a new method based on statistics from population genetics, signature variability analysis (SVA), that elucidates both the diversity of tumorcausing processes and the heterogeneity of population carcinogen exposure. When we use SVA to re-analyze four prominent studies commonly cited as evidence of nonmutagenic carcinogens, we find that tumors induced by environmental carcinogens do possess mutational signature patterns that distinguish them from spontaneous tumors, even if a specific mutational signature cannot be detected. We find that, across cancers, organs, and model organisms, carcinogen exposure generally increases both the diversity of mutational signatures within a tumor and the homogeneity of signature activity across subjects. Importantly, we show that this increase in signature diversity, far from being a background effect, is associated with the geographic incidence of cancer and can facilitate the acquisition of cancer driver mutations. Our results both encourage a re-examination of the genomic impact of numerous substances and introduce new tools for the analysis of the genomic effects of other substances, potentially influencing carcinogen classifications and cancer prevention policies.