When a coal mine fire occurs, the rescue of trapped mine workers is difficult, because some efficient fire-fighting measures (such as nitrogen injection) cannot be implemented. The development and severity of a fire is difficult to determine, and the integrity of the auxiliary ventilation system may be unknown. These situations pose additional challenges to the trapped mine workers and increase the difficulty level for the rescue team. A novel method using the tracer gas technique is developed to remotely gather information to determine the location and severity of the fire. Laboratory experiments were conducted to understand the tracer gas distribution characteristics when the fire causes damage to the ventilation duct. A mathematical model was developed that uses the tracer gas concentration curve to determine the location and severity of the fire. The model combines the Expectation Maximization (EM) algorithm with the Gaussian Mixture (GM) model. Validated using experimental data, it is demonstrated that the method can determine the fire location with low error. The information collected using this method reflects basic living environmental conditions of the trapped mine workers and provides important input for quickly formulating an effective fire rescue plan that saves lives.