Cancer drivers require statistical modeling to distinguish them from passenger events， which accumulate during tumorigenesis but provide no fitness advantage to cancer cells. The discovery of driver genes and mutations relies on the assumption that exact positional recurrence is unlikely by chance; thus， the precise sharing of mutations across patients identifies drivers. Examining the mutation landscape in cancer genomes， we found that many recurrent cancer mutations previously designated as drivers are likely passengers. Our integrated bioinformatic and biochemical analyses revealed that these passenger hotspot mutations arise from the preference of APOBEC3A， a cytidine deaminase， for DNA stem-loops. Conversely， recurrent APOBEC-signature mutations not in stem-loops are enriched in well-characterized driver genes and may predict new drivers. This demonstrates that mesoscale genomic features need to be integrated into computational models aimed at identifying mutations linked to diseases.