Abstract: With the development of mineral resources exploration, the seismic exploration and acquisition environment is more and more complex, and the received reflection seismic records often contain a lot of interference, which results in low signal-to-noise ratio, complex appearance and difficult to distinguish the effective information of field seismic data, which makes it difficult to extract and utilize the effective information in seismic records. Traditional numerical simulation records generally do not have a random interference noise background containing various causes. Although the processing effect of theoretical records is good, it is difficult to process the actual seismic data with low signal-to-noise ratio. Up to now, there are few forming theories for seismic random noise simulation methods, and there are some idealizations and limitations for the understanding of random noise characteristics of seismic exploration. In noise simulation, many studies add white noise which can be generally eliminated in subsequent superposition processing, and cannot be close to the actual data processing. Based on the classical decomposition theory of random functions and Marmousi model records, the additive random interference is calculated according to the coherence characteristics of the CMP gathers, which is more similar to the interference background of actual seismic data because of its similar coherence with CMP gathers records. This method can guide the design of field observation system for seismic exploration, post-processing and interpretation of seismic data.