When particle swarm optimization (PSO) is adopted to optimize antenna array pattern, the optimizer’s convergence can be greatly improved by the efficient estimations of the optimum particle’s initial values, and some global optimal solutions better than usual can be obtained. By using the array’s Taylor weighting to estimate the optimum particle’s initial values, a new method is presented to improve PSO optimizer’s convergence. Comparisons are made between three methods of estimations, which are random weights, Taylor weights and analytic weights respectively, it can be said that Taylor weights and analytic weights are both effective to improve the PSO optimizer convergence relative to the random weights, and they both have similar performance. As the analytic weights only can be solved by performing large-scale matrix’s generalized inversion which is usually difficult and hardly afforded by common PC, while Taylor weights are relatively simple and easy to be calculated, the practicability of the latter is far better than the former.