Due to Globalization economies have become more integrated than ever thereby making exchange rate more important for both policy makers and businesses. Pakistan, being an emerging economy, is quite attractive place for foreign portfolio investment. Hence a good understanding and modeling of exchange rate is required for forecasting, in order to make wise decisions regarding international business and economic policies. The general consensus, in this regard, however, is that exchange rate is random walk process. The present study contributes to the ongoing debate regarding the possibility of correctly forecasting future exchange rate movements. The econometric evidence resulting from this kind of study can suggest which model should be adopted in order to achieve a better forecasting performance contrary to general perception that exchange rate is a random walk process. Our results suggest that exchange rate can reasonably be forecasted if structural breaks are properly modeled. We have used linear and nonlinear model in the form of autoregressive process and threshold autoregressive process, respectively. Quarterly data of Pakistan over the period 1981:Q4 -2011:Q1 have been gathered on nominal exchange rate, real exchange rate, and real effective exchange rate for testing efficiency of linear and nonlinear models to forecast the exchange rates. Our results reveal that nonlinear model can better forecast the nominal and real exchange rates, but in case of real effective exchange rate, linear model is found better. Supervised by: Dr. Wasim Shahid Malik