In this study, the optimal m-step set points of temperature that minimize the water loss of the fruit were determined using a decision system and applied to a real system. In the decision system, the dynamic change in the rate of the water loss, as affected by temperature, was first identified using neural networks, and then the optimal m-step set points were sought through simulation using genetic algorithms. Several types of optimal values were obtained under different constraints of the temperature and evaluation lengths of the control process. A temperature operation first rising to the highest level and then dropping to the lowest level in the given range was effective to reduce the water loss than that keeping it constant at the lowest level.