Within the scope of visible spectrum, the color of high-temperature object changes with the temperature. A method of temperature measurement using image color based on neural network is proposed. Hue and saturation of the HSV model are used as the pattern character vectors. BP neural network is used to approximate the nonlinear relationship between the color and the temperature of the high-temperature object. The experimental result shows that the accuracy is within acceptable scope, and the result can't be effected by the environment, so the method is feasible.