Numerous studies have linked gut microbiome alterations to colorectal cancer (CRC), but limited sample sizes and study heterogeneity have hampered cross-study comparisons and subgroup analyses. Here, we present a comprehensive single-disease gut microbiome meta-analysis based on consistently re-computed and re-analyzed shotgun and amplicon sequencing profiles (n = 6,779 samples, 27 studies). Association and machine-learning analyses delineate CRC microbiome signatures, which are robustly generalizable across studies and sequencing approaches and nearly identical between early- and late-onset cases. Meta-analysis of the tumor-resident microbiome reveals characteristic tumor-enriched microbes in concordance with fecal signatures that are clearly detectable in early-stage tumors, although their detection in feces becomes moderately higher in late-stage and distal tumors, possibly due to dilution effects in stool. The unified fecal CRC signature inversely associates with dietary fiber intake and is modifiable by dietary interventions. Finally, genome-resolved functional analysis reveals variation in virulence factor carriage and geographic enrichment across Fusobacterium subspecies.