Safety signs play a very important role in people’s evacuation during emergencies. In order to explore the appropriate color for subway safety signs, four safety signs of different color combinations are designed, and the virtual reality, eye-tracking technology, and physiological indicator measurement are used in a virtual subway fire escape experiment. A total of 96 participants with equal distribution in gender and four different color combination groups were recruited. Participants’ eye-tracking and physiological data (heart rate, skin conductance) were real-time recorded through ErgoLAB V3.0 in the whole experiment. The relationship between Color_of_safety_sign and escape performance, eye-tracking indicators, and physiological indicators is discussed respectively through SPSS. The results show that “Green and black” group has the best evacuation escape performance, low cognitive load, high search efficiency on safety signs, and the highest stress level and immersion and “Green and black” can be the most appropriate color for safety sign. This research is of certain significance for improving the function of subway fire-fighting infrastructure and the resilience of the metro system. Moreover, it can provide references and advice on risk management, emergency evacuation, and so on.