The current expansion of RNA epitranscriptomics calls for direct and high-precision mapping tools to characterize the intrinsically low abundant RNA modifications. Here, we developed a strategy, termed Graft-seq, which harnesses specific enzymatic and chemical reactions on an RNA modification site to covalently graft a known RNA branch and further utilizes the branch-to-main-chain or main-chain-to-branch landing/jumping site signal during reverse transcription to determine the locations of RNA modifications at single-base resolution. We developed a matched bioinformatics analysis pipeline for Graft-seq and successfully mapped internal N(6)-methyladenosine (m(6)A) and cap N(6),2'-O-dimethyladenosine (m(6)Am), as well as nicotinamide adenine dinucleotide (NAD) on transcriptome-wide mRNAs and/or nuclear non-coding RNAs in different cell lines. Paralleled comparisons of Graft-seq with available techniques confirmed the effectiveness of the RNA grafting strategy. Graft-seq represents a direct and enrichment-free technique for characterizing RNA modifications at base resolution and offers the potential to discover new RNA modifications and RNA-RNA interactions.