An adaptive polarization mode dispersion (PMD) compensation experiment for 10 Gbit/s nonreturn-to-zero (NRZ) and return-to-zero (RZ) optical fiber communication systems with a two-stage compensator is successfully realized, in which degree of polarization (DOP) feedback signal is used as PMD monitor. The algorithm so-called particle swarm optimization (PSO) is used in optimum searching and tracking process. It was proved that in the optimum searching process, the PSO algorithm has the characteristics of rapid convergence to the global optimum rather than being trapped in local sub-optima and robust to noise. In the tracking process, the algorithm can track the random changed PMD with fast speed. The compensator shows its ability to compensating the first- and second-order PMD at same time. After compensation the eye diagrams can be well recovered both for NRZ and RZ formats. The searching process is finished within several hundreds of milliseconds and the response time of recovering to optimum for a sudden PMD disturbance is less than 20 milliseconds. The quasi-real time adaptive PMD compensation is realized.