A pairs trading strategy (PTS) forms a market-neutral portfolio whose value moves back and forth around a certain price level. An investor can long (short) the portfolio when its price moves below (above) the price level and cash out when the portfolio value converges back to earn the price difference. The profit for each successful trading is relatively small, so transaction costs and structural changes (that invalidate market-neutral property) could significantly erode the profits. This paper proposes three improvement methods to reduce the costs and stabilize aggregated profits. First, we change the open and close thresholds to increase the number of transactions; this would increase and stabilize the aggregated profits due to the law of large numbers. Second, we derive the expected return for each trading before opening the portfolio and execute the trading only when the expected return exceed the transaction cost. Third, we detect the structural change with our revised (statistical) tests to close the position in advance to reduce losses. Empirical studies show that our three methods can be simultaneously adopted to improve trading performance significantly.