The invention discloses a medical data security sharing method based on a block chain and federated learning. The data applicant can use the data after being authorized on the chain of the data provider; the data fingerprint carries out hash abstract chaining on authorized data, the problem that the authorized data is maliciously tampered to cause data inconsistency is prevented, the use right ofthe original data is shared in the whole process, a data user cannot directly obtain the data, and the value of the data can be mined only through federated learning. In each round of iterative computation of federated learning, asset chaining is carried out on model parameters and aggregation results, and credible traceability of federated learning computation can be achieved. Each step of operation in the data sharing process is subjected to related auditing by a supervisor, such as identity auditing, data checking, transaction detail auditing and the like. According to the invention, aggregation calculation is carried out without the help of a central server, decentralized federated learning is realized, aggregation calculation is realized through an intelligent contract, and maliciousaggregation calculation results received by each node due to malicious control of the central server are avoided.