Objective: To establish an artificial neural network for predicting retention rate of total polysaccharides in fibrous rhizome herbs by ultrafiltration technology. Method: Taking pore size of inorganic ceramic membrane

pressure and temperature as input parameters

retention rate of total polysaccharides in enzymatic hydrolysate of Hedysari Radix under different ultrafiltration conditions as output parameter

BP neural network was established after network parameters optimized by Levenberg-Marquardt method.Then performance and applicability of model were evaluated. Result: Topological structure of BP neural network was 3-6-1.Mean prediction error

mean absolute error and mean error rate of BP neural network for retention rate of total polysaccharides in Hedysari Radix was 0.10%

0.98% and 1.55%

respectively.Mean error rate for retention rate of total polysaccharides in Astragali Radix was 2.77%. Conclusion: Accuracy and applicability of BP neural network is good enough to predict retention rate of total polysaccharides in fibrous rhizome herbs by ultrafiltration technology.