TCP congestion control mechanism, being based on the presupposition that network congestion is the only reason for data loss, is suitable for wired environments rather than for wireless environments with characteristics of higher bit error rate, lower bandwidth, longer delay, frequent mobility and so on, where link corruption can also cause data loss. To avoid unnecessary decrease of packet sending rate resulting in lower throughput and longer delay, some improved TCP congestion mechanisms have been presented. However, one of the disadvantages of these mechanisms is that a great deal of unnecessary packets lost by corruption can be caused while heavy data loss rate by corruption lasts short time. As well known, energy consumption of mobile hosts is one of the most important factors for wireless communication. How to save energy of mobile host has become an important task for wireless communication. One of the methods is to improve reliability to decrease overheads resulting from re-transmitting lost data. Obviously, it is necessary to decrease data sending rate, but how does it do? This paper answers the question, and then presents an improved TCP congestion control mechanism, namely TCP_IR (TCP for Improving Reliability) whose implementations are based on two algorithms. Algorithm I-based TCP_IR is driven by data sending periodic, while algorithm II-based TCP_IR is driven by the event of data lost by corruption. They can be very easily implemented with fewer overheads and can effectively improve TCP reliability with light influence to throughput and delay. The paper’s contribution is to give valuable references to further study TCP in wireless networks.