This paper proposes StepProof, a novel automatic formalization method that converts natural language proofs into formal proofs using large-scale language models (LLMs). Unlike existing methods that only verify complete proofs, StepProof decomposes proofs into several verifiable sub-proofs, enabling sentence-level verification. Experimental results show that StepProof significantly improves the proof success rate and efficiency over existing methods, and that the performance can be further improved by adding some manual adjustments to natural language proofs.