Abstract: Back propagation neural network (BPNN) was applied to simulate and predict the extraction of camellia seed oil by supercritical CO2. Neural network's structure, training and optimization of learning algorithms and hidden layer structure were investigated. Under different conditions of feed particle's mean diameter (0.215-0.625 mm), pressure (30-35 MPa), temperature (35-50℃) and CO2 flowrate (20-25 L/h), the yields of camellia seed oil were predicted by BPNN. The results show that L-M algorithm is the optimum learning algorithm of the BPNN, and the BPNN with 5/8/1 network structure has good simulating performance. The model predictions are in good agreement with the experimental data, and relative errors among most of the predictions and the corresponding experimental data are less than 3.0%. The BPNN developed can be used in simulation of this process.