Abstract: Heavy involvement of artificial subjective factors in dosing agents in flotation process generally leads to fluctuation of ash of flotation products and high consumption of agents. To address this issue, two agent dosage prediction models based respectively on BP neural network and GRNN neural network are developed and the hardware system is designed. As revealed through verification test on the dosing amounts of collector and frother predicted by the two models, the BF neural network-based model can offer a relatively better prediction result and is more suitable for use in industrial flotation process.