Estimating dynamic roll motion under fishing operation conditions is one of the unresolved problems with respect to the capsize accident of fishing vessels. In addition to potential non-linear response of the roll motion to the wave, unexpected factors occurred by fishing operations make the roll motion complicated. Considering the estimation for the dynamic phenomena generating on the fishing vessels, the numerical simulation based on non-deterministic approaches is one of the effective methods. Among them, it is well-known that a Neural Network model is a practical methodology. Based on the tank tests, the roll response models in fishing operation conditions were constructed. Appropriate estimates for the actual roll motion with a full-scale showed the effectiveness of the Neural Network model. Furthermore, as the model forecasts properly the critical condition occurring for the shipping water, it is suggested that the model is also effective utilizing as the forecasting model to avoid the fishing disaster.