In this paper, we try to explore the foreign exchange market microstructure with all transaction records in a day. Our data set，“EBS Market Data”，contains all the transaction in a day. EBS Market Data
records all the transaction for 24 hours starting from 22:00:00 in previous day to next day. We apply Autoregressive Conditional Duration model for uneven spaced tick by tick foreign exchange rate. This data
set enables us to use limit order data, market order data and exit information
of limit order individually. Fact findings are as follows. New York market is the busiest no matter what kind of currencies dealers trade and regardless of the time of the day. Around half of limit orders of three currencies are cancelled. With ACD model, result suggests strong evidence that there exists duration persistency in all data set. Volume has positive and volatility has negative impact. Larger volume order shorten the duration, and higher volatility induce more trades.