<p>To handle the ill-posed inverse problem in the measurement of particle size distribution (PSD) of polydisperse particle system using dynamic light scattering (DLS), a hybrid method named as CONTINfit-GRNN is proposed in this paper. The CONTINfit-GRNN is constructed by taking advantages of the constrained regularization method (CONTIN), the nonlinear least squares fitting strategy and a general regression neural network (GRNN). Simulation data for the electric field autocorrelation function (ACF) and fitted data for CONTIN are used to generate the training set, and the optimal smoothing parameter, which adjusts the generalization ability of GRNN, is determined by minimizing the deviation of the distribution. To validate the capability of this method, CONTINfit-GRNN is used to estimate both the unimodal and bimodal PSDs. Simulation results show that CONTINfit-GRNN achieves higher accuracy than fitting data of CONTIN in both narrow and broad distributions, and also shows high stability in the inversion of PSD as the noise level increases.</p>