The feedback control algorithm is the key integral part of an adaptive polarization mode dispersion (PMD) compensator. The particle swarm optimization (PSO) algorithm was introduced into the adaptive PMD compensation system, which showed the merits of rapid convergence to the global optimum, not being trapped in local sub-optima, robust to noise and capability of multi-degree of freedom control. The performance of two kinds of PSO algorithm, global version of PSO (GPSO) and local version of PSO (LPSO), was analyzed. The three kinds of topologic structures with which the PSO algorithms had 100% success probability were given. The experimental performances of GPSO and LPSO in compensating the first-order and second-order PMD showed that GPSO and LPSO were both qualitied when they are used in compensating the first-order PMD, but LPSO behaved better than GPSO when they are used in compensating the second-order PMD.