In this paper, in case of fire in a tunnel, I performed the real test for the propagation time of toxic substances and a three-dimensional numerical analysis, and compared each results. A smoke generating device was produced for the real test, capable of 8.8 kW of the heat source and 90 l/h of smoke generation. The target tunnel is 570m in length, 64.6m2 in cross-sectional area, and has a slope of approximately 2%. To reduce the impact of the draft at the exits of the tunnel, the blocking curtains was installed. The smoke spread, speed and temperature were measured as a 20m interval from the location of the smoke device every 20 seconds. After 60 seconds, smoke reached 20m away from the device, after the lapse of 140 seconds, 180 seconds and 260 seconds, respectively 40m, 60m and 80m position was reached. The results of three-dimensional numerical analysis were over-predicted about 26.3 to 49.5%, when the tunnel wall was treated as a smooth surface. In addition, in the case of the tunnel wall with a rough surface with a certain roughness, as an 2.7∼17.6% error, it showed more closely on the results of the experiment. Thus, when performing three-dimensional numerical analysis on the behavior of pollutants in the tunnel, it shall be judged the consideration of the wall roughness was required.