Abstract: Reduction roasting magnetic separation process can effectively extract nickel, iron and other valuable metals from laterite nickel ore. Due to the multiple factors existing in the process of reduction roasting magnetic separation of laterite nickel ore, the industrial indicators are unstable. In order to further improve the effect of reduction roasting magnetic separation process in laterite nickel ore, the factors of reducing agent dosage, roasting temperature, material thickness, roasting time and magnetic field intensity were optimized with a nickel ore in Qinghai as raw material by combining orthogonal experiment and BP neural network. The results showed that the optimized experimental conditions by BP neural network model are as follows: dosage of reducing agent 9.5%, roasting temperature 1 070 ℃, layer thickness 10.0 mm, roasting time 65 min and magnetic field strength 2.5 kA·m^-1. Under these conditions, a rough nickel concentrate with a yield of 30.29% can be obtained, which is 2.83% higher than the yield of nickel rough concentrate obtained by using the optimal factor combination conditions of the orthogonal test.