The present invention discloses a floating point data protection method during a data link transmission process, comprising: abstracting, from floating point data, a representation form of a single-precision floating point number and a double-precision floating point number during a transmission process; performing a bit field operation on the floating point data to construct a data structure thereof; and then excluding abnormal numbers and unreasonable numbers thereof to obtain a correct floating point number. Therefore, the problem of a computer crash or a wrong result caused by an error in floating point data during a multi-byte data link transmission process is solved, making the embedded system have higher security and reliability.