Quantitative structure-mobility relationship (QSMR) models were developed between the structures of flavonoids and their eletrophoretic mobilities in micellar electrokinetic capillary chromatography. Molecular descriptors calculated from structure alone are used to represent molecular structures, moreover, Nt was defined by ourselves. Multiple linear regression and radial basis function neural networks (RBFNNs) are utilized to construct the linear and nonlinear prediction model, respectively. The optimal QSMR model developed was based on a 3-10-1 RBFNNs architecture. The root mean square errors in mobilities predictions for the data set was 0.1083 mobility unit (10(-4) cm2 V(-1) s(-1)). The prediction results were in good agreement with the experimental values.