Mixed Weibull distributions are widely used in the reliability data analysis of complex systems with multiple failure modes. Since the distribution functions contain much more parameters, the parameter estimation of the mixed Weibull distribution is more complicated than the single Weibull distribution. A least square method for parameter estimations of mixed Weibull distributions is proposed by using a genetic algorithm as the optimization method. The minimum residual sum of square is taken as the optimization objective. The ranges of each parameter are taken as the constraints. And a nonlinear least square model of the mixed Weibull distribution is developed. The traditional genetic algorithm has been improved by altering the upper and lower bounds, introducing the penalty factor and preserving the optimal individual. And the nonlinear least squares model is solved based on the improved genetic algorithm. A case study is presented to illustrate the effectiveness of our proposed method. The maximum deviation and the normalized root mean square error obtained by our method are reduced by 0.028 4 and 0.032 8 compared with those of the graph estimation method, and reduced by 0.000 8 and 0.003 61 compared with those of the maximum likelihood estimation method.