During the voyage, the fishing vessel is in a potential threat because of its own structure or the influence of sea surface wind and waves. In order to study the risk of fishing vessels in the marine environment, based on the BP neural network algorithm, the fishing boats early warning model which is composed of 6 early warning indicators : fishing vessel tonnage, engine power, material, fishing vessels age, sea breeze level, wave level, were evaluated and then the sea operations risk level for fishing vessels were finally determined. 400 fishing vessel accident cases were selected to develop the risk early warning model  and  the model was verified through classification of multiple levels for the training samples. The results of early warning and the actual results of statistical calculation showed, the correct rate remained at 79.76%-83.62%, in which when the training sample number was 0.75 times as the number of test samples, the accuracy of the model is highest. In conclusion, the assessment results of fishing vessel risk early warning model based on BP neural network was basically consistent with the actual condition of accident, which could provide guarantee for safe navigation .