Colorectal cancer (CRC) is the third most common cancer, and it can be prevented by performing early screening. As a hallmark of cancer, the human microbiome plays important roles in the occurrence and development of CRC. Recently, the blood microbiome has been proposed as an effective diagnostic tool for various diseases, yet its performance on CRC deserves further exploration. In this study, 133 human feces and 120 blood samples are collected, including healthy individuals, adenoma patients, and CRC patients. The blood cfDNA and fecal genome are subjected to shotgun metagenome sequencing. After removing human sequences, the microbial sequences in blood are analyzed. Based on the differential microbes and functions, random forest (RF) models are constructed for adenoma and CRC diagnosis. The results show that alterations of blood microbial signatures can be captured under low coverage (even at 3×). RF diagnostic models based on blood microbial markers achieve high area under the curve (AUC) values for adenoma patients (0.8849) and CRC patients (0.9824). When the fragmentation pattern is combined with microbial and KEGG markers, higher AUC values are obtained. Furthermore, compared to the blood microbiome, the fecal microbiome shows a different community composition, whereas their changes in KEGG pathways are similar. Pathogenic bacteria