It is difficult to measure the surface temperature of iron ore directly. A method is put forward to handle this problem by using softsensing technique. This online measurement method is used to replace the Lagrange interpolation offline method to estimate surface temperature. The method used LM optimum algorithm to build up ANN soft sensor model combined with offline learning neural network to establish the correlation between input variables and target variables, to achieve the surface temperature on line detection. The results of simulation and experimentation indicate that the method is reasonable and feasible.