Abstract: Aimed at the characteristics of marine nuclear power plant that fault and corresponding sign are various, and in order to overcome inhere problem of traditional genetic algorithm, improved genetic algorithm and probability causal model were integrated to establish fault diagnosis system. The system is made up of data collection module and fault diagnosis module with high real time and speed. To validate the system, stream generator was diagnosed, and the computation result is in agreement with conclusion of some literature. At last, the comparison with tradition arithmetic was made. The results indicate that with improved genetic algorithm, mistaken diagnosis is improved effectively, and accuracy of fault diagnosis is enhanced.