Alternative polyadenylation (APA) is emerging as a major mechanism of post-transcriptional regulation. APA can impact the development and progression of cancer, suggesting that the genetic determinants of APA might play an important role in regulating cancer risk. Here, we depicted a pan-cancer atlas of human APA quantitative trait loci (apaQTLs), containing approximately 0.7 million apaQTLs across 32 cancer types. Systematic multi-omics analyses indicated that cancer apaQTLs could contribute to APA regulation by altering poly(A) motifs, RNA binding proteins (RBP), and chromatin regulatory elements and were preferentially enriched in GWAS-identified cancer susceptibility loci. Moreover, apaQTL-related genes (aGenes) were broadly related to cancer signaling pathways, high mutational burden, immune infiltration, and drug response, implicating their potential as therapeutic targets. Furthermore, apaQTLs were mapped in Chinese colorectal cancer (CRC) tumor tissues and then screened for functional apaQTLs associated with CRC risk in 17,789 cases and 19,951 controls using GWAS-ChIP data, with independent validation in a large-scale population consisting of 6,024 cases and 10,022 controls. A multi-ancestry-associated apaQTL variant rs1020670 with a C>G change in DNM1L was identified, and the G allele contributed to an increased risk of CRC. Mechanistically, the risk variant promoted aberrant APA and facilitated higher usage of DNM1L proximal poly(A) sites mediated by the RBP CSTF2T, which led to higher expression of DNM1L with a short 3'UTR. This stabilized DNM1L to upregulate its expression, provoking CRC cell proliferation. Collectively, these findings generate a resource for understanding APA regulation and the genetic basis of human cancers, providing insights into cancer etiology.