In our study, the non-linear regression model and artificial neural networks (ANNs) were used to optimise the preparation of the loading norcantharidin chitosan nanoparticles (NPs) by ionic cross-linkage. Two major indexes, the particle size and the entrapment efficiency of the drug vehicles were synchronously optimised according to the normalised value calculated referring to the weights of the indexes and factors. For the purpose, a multiple regression model was constructed for fitting several preparation factors, including the low molecular weight of chitosan (LCS), sodium tripolyphosphate (TPP) concentrations and the temperature of the ionic cross-linkage reaction. Each of the level values in the factors was arranged using the L 9 (3 4 ) table and their linear weighted sum of the normalised value was taken as optimised object. A back-propagation (BP) network (3 Â 7 Â 2) in ANNs was created and trained for further checking the optimal results and the trained network was applied to simulate the experiment system and screen the optimal conditions. Finally, when the weights of temperature, particle size and entrapment efficiency were 0.1, 0.4 and 0.5, respectively, the best preparation condition of NPs was obtained as 131 AE 7 nm of particle size and 45.12% of entrapment efficiency at 40 C.