Research purposes: Figuring out the characteristics of early smoke flow from a running train fire is the foundation of fire detection, location and warning. Based on the numerical simulation model of train fire, six different-location-fire models of roof, bottom, front, middle and rear of 120 km/h running train are used to study the variation of air temperature and smoke concentration at different times.Research conclusions: (1) The smoke is asymmetrically distributed, and tends to spread to the rear, it can not detect the changes in temperature and smoke concentration from 1 m distance ahead of the fire. (2) The real-time monitoring values of temperature and smoke concentration are negatively correlated with the distance from fire source, and the air temperature of an under -train-fire is 100% higher than that of a fire on the roof, and the smoke concentration is 500% higher. (3) The relative velocity has little influence on the absolute temperature ratio. The absolute temperature ratio is greater than 0.45 (1 m away from the fire) and 0.75 (3m away from the fire) whether the relative velocity is 5 m/s or 10 m/s. (4) The research results could provide references for fire protection technical standards and the design of smoke management systems. © 2024 Editorial Department of Journal of Railway Engineering Society. All rights reserved.