This paper empirically tests the purchasing power parity (PPP) using panel unit root tests. We employ a battery of panel unit root tests: LM-bar statistic [Testing for unit roots in heterogeneous panels, Working paper, University of Cambridge] is employed to account for serially correlated errors. The statistic proposed by Breitung [Adv. Econom. 15 (2000) 161.] and the KPSS-based statistic of Hadri [Econ. J. 3 (2000) 148.] are also used. In addition, we also employ a SUR estimator to account for possible cross-sectional effect. Data of 45 economies from 1980 to 1999 are used to test the PPP hypothesis. We find that these estimators tend to get supportive results when the data frequency becomes lower, which substantially characterizes the long-run property of the PPP hypothesis.