A modeling of total evaluation process of sake(Ginjo) was studied using a fuzzy neural network(FNN). Total evaluation of 61 Ginjo samples was estimated from each data set of 7 sensory evaluations. The values of performance index, J, based on the errors between actual and estimated values in FNN model were used for the evaluation of model. In FNN model with all 7 input variables, J value was 0.025 and it was almost similar to that of NN model as reported previously. The FNN model with 3 variables (color, flavor base and aging) selected previously was also tested, and J value (0.023) was also almost similar to that of NN model (J=0.024). To optimize the input variables and the number of membership functions, Parameter Increasing Method (PIM) was applied to the part of premise in FNN model. The FNN model obtained was constructed with 2 membership functions for color, 3 for flavor top, 2 for flavor base and 3 for hard-soft, and J value of 0.013 was fairy small. From analysis of connection weight of the FNN obtained, the acquired rules were easily described in the form of IF-THEN rule. Extraordinary flavor was added as a new input variable to above 3 FNNs. J values in all models decreased furthermore and especially the FNN with 3 variables and extraordinary flavor was found to show the lowest J value (0.010). It was concluded that 4 variables (color, flavor top, aging, extraordinary flavor) were important in the total evaluation of Ginjo. The results suggest that the fuzzy modeling using a FNN is effective on the analysis of sensory evaluation process.