This paper examines, using various models including a non-linear one, that the Balassa-Samuelson (BS) effect can account for the persistence of deviations from PPP in the long-run movements of won/dollar and won/yen real exchange rates. In test for PPP hypothesis that incorporates the BS effect, using the generalized Johansen' cointegration method, it is found that a cointegration relationship exists between each of won/dollar and won/yen real exchange rate and the productivity variables of two countries. And in test for PPP hypothesis that incorporates other fundamentals such as cumulative current account balance, foreign exchange reserve, terms of trade as well as productivity differentials, using a behavioral equilibrium exchange rate approach, it is found that a cointegration relationship exists between each of won/dollar and won/yen real exchange rate and all of these fundamentals. However, the plus sign of the estimated coefficient of the productivity differentials variable, which means that domestic productivity improvement produces increase in each of won/dollar and won/yen real exchange rate is not coincident with the result that the BS effect expects theoretically. Finally, in test for PPP hypothesis that incorporates the BS effect, using a non-linear STAR model, it is found that the adjustment process in case of won/dollar real exchange rate from the long-run equilibrium level can be adequately explained by a non-linear LSTAR model. But, the evidence of diagnostic statistics, which shows the existence of autocorrelation of the residuals in most of lags, might suggest the inadequacy of LSTAR model specification.