Circulating tumor DNA (ctDNA) has emerged as a tumor-specific biomarker for the early detection of various cancers. To date, several techniques have been devised to enrich the extremely small amounts of ctDNA present in plasma, but they are still insufficient for cancer diagnosis, especially at the early stage. Here, we developed a novel method, CUT (CRISPR-mediated, Ultrasensitive detection of Target DNA)-PCR, which uses CRISPR endonucleases to enrich and detect the extremely small amounts of tumor DNA fragments among the much more abundant wild-type DNA fragments by specifically eliminating the wild-type sequences. We computed that by using various orthologonal CRISPR endonucleases such as SpCas9 and FnCpf1, the CUT-PCR method would be applicable to 80% of known cancer-linked substitution mutations registered in the COSMIC database. We further verified that CUT-PCR together with targeted deep sequencing enables detection of a broad range of oncogenes with high sensitivity (<0.01%) and accuracy, which is superior to conventional targeted deep sequencing. In the end, we successfully applied CUT-PCR to detect sequences with oncogenic mutations in the ctDNA of colorectal cancer patients' blood, suggesting that our technique could be adopted for diagnosing various types of cancer at early stages.