Abstract: In criminal justice activities, big data has emerged as a way of proof that is different from the traditional proof. Through the observational analysis of the applicable cases in China's current judicial practice, it can be summarized that the proof role of big data in the field of criminal justice is mainly manifested in five aspects: proving the identity of the subject involved in the case, the relevance of the information carrier, the amount of the massive information involved in the case, the criminal behavior, and the characteristic information in the criminal composition. Compared with traditional judicial proof, the most distinctive features of big data proof are intelligence, modeling and special relevance. To understand its inner principle and operation mechanism, we can adopt a kind of elemental thinking of “massive information + data processing + conclusion”. Based on the judicial application and the inner mechanism of the theoretical review of big data proof, we should firstly establish the legal attribute specification of the evidence-based conclusion of big data proof that can be used as the basis of the case, secondly clarify its factual determination principle as the opinion of the machine, and finally try to establish the paradigm of the binary differentiation of the general and case-specific review process based on the stability of the algorithm application scenarios.