The relationship of distribution coefficients of decay linewidth and exponential terms is linear in the autocorrelation function of dynamic light scattering signal. When the linear estimation and nonlinear least squares method are combined,double exponential algorithm is improved. By improving algorithm the initialization parameters are reduced and four parameters optimization problem are changed into two parameters optimization problem so that fitness results is more stability,higher accuracy. Therefore the inverted results of particle size are more accurate. The autocorrelation function of light scattering signal of 100 nm and 1000 nm with two-peak distribution particles is inverted. Inverted error of improving algorithm for noise-free autocorrelation function is zero,which is superior to double exponential algorithm of at least 1.708%. For inversion of noise autocorrelation function,improving algorithm reduces the error of 0.558%-5.738%. The results show that the improving algorithm is better than double exponential algorithm in the capability of optimization and tolerance of noises.