Abstract: In order to solve the problem of high alarming error rate of traditional fire pre-warning algorithms of belt conveyor which alarm according to single characteristic signal, a fire pre-warning algorithm of belt conveyor was proposed which combines the traditional fire alarming algorithms with neural network. The individual characteristic variables extracted by the traditional algorithms are used as inputs of neural network, and the neural network is trained by use of sample data measured in laboratory. The output of the neural network is the final output of the fire pre-warning algorithm. The experimental result showed that the algorithm can effectively identify fire pre-alarming status of the sample data with low error rate.