LED lights are expected to improve the visibility of emergency vehicles. We propose a two-stage method of evaluating the flash patterns of LED warning lights. In the first stage, we identify the visibility of many kinds of flash patterns using the mental measure of "distinctness" from Scheffe's paired comparison. In the second stage, we select a few kinds of flash patterns with higher distinctness out of patterns used in the first stage, and get the reaction time to their patterns with high accuracy. We conducted experiments on 3DCG simulation by creating CG animations as stimuli which LED Lights flashed in various patterns. The results of the two-stage evaluation using 11 flash patterns indicated that the mental measure of distinctness of the flash patterns of LED is correlated to the reaction time. We show a possibility that the most-visible flash pattern can be designed with this method.