This paper investigates the empirical relation between order imbalance and intraday NTD/USD exchange rate dynamics. From market microstructure arguments, order imbalance could convey private information about macroeconomic fundamentals, and thus have both explanatory and forecasting power for exchange rates. Using one year of high frequency data, our time-series regression model demonstrates that interbank order imbalances have substantial explanatory power for concurrent exchange rate returns both on the daily and intraday basis. More importantly, we find that lagged-one order imbalances have a predictive negative effect on current returns. Further, we trace the predictability of returns using order imbalances over various intervals to investigate the intraday market efficiency. We show the weak-form efficiency appears to prevail over intervals of from fifteen to sixty minutes in the NTD/USD exchange rate market. Then, based on the predictive negative relation of lagged order imbalances and current returns, we develop a contrarian trading strategy with different order imbalance truncations both on the daily and intraday basis. We find no matter what kinds of scenarios we choose, order imbalance trading strategy yields a positive return, and the 90% truncation strategy consistently dominates the benchmark. Furthermore, according to the nested causality analysis, we find that a unidirectional relation from order imbalance to return on a daily basis, while a contemporaneous relation between return and order imbalance on an intraday basis. This result could explain why the performance of our daily strategies could dominate that of the intraday strategies. Besides, we propose a GARCH (1,1) model by simultaneously incorporating order imbalance in the conditional mean and variance equations to model the time-varying NTD/USD exchange rate. We find the price impact of interbank order imbalance decrease after considering the volatility effect. This finding is consistent with interpreting estimates of current relation between high frequency returns and order imbalance as a measure of liquidity (Brenan and Subrahmanyam, 1996).