Abstract: In order to obtain optimum condition for soluble dietary fiber from orange peel extracted by fluid CO2, BP Neural Network (BPNN) combined with response surface methodology (RSM) was put forward as a new method to analyze and process test data.The response surface test data could be used by applying the self-learning ability of BPNN.With the help of BPNN, which simulated, evaluated and optimized, the result of experiment showed that BPNN had the less errors than RSM.The optimum conditions for soluble dietary fiber from orange peel were as follows:temperature 91.5℃, pH1.60, homogenized pressure 30.5MPa, and time 2.14h.The experiment was accomplished under optimum conditions, the yield of soluble dietary fiber was 28.79%.In conclusion, BPNN provided a technonical basis for industrialization production of soluble dietary fiber from orange peel.