As a well-known issue, the performance of transmission control protocol (TCP) is affected by its loss recovery mechanism working based on two algorithms of fast retransmit and fast recovery. In particular, the fast recovery algorithm has been modified for improvement and become a basis on which each TCP implementation is differentiated such as TCP Tahoe, Reno, NewReno, and selective acknowledgement (SACK) option. Despite the recent improvement achieved in TCP NewReno and SACK option, TCP still has a problem that it cannot avoid retransmission timeout (RTO) when a retransmitted packet is lost. Therefore, in this paper, we introduce a simple algorithm for detecting a lost retransmission that can be applied to TCP implementations that do not use SACK option such as TCP Reno and NewReno. Using ns simulations, we show the microscopic behaviors of the lost retransmission detection algorithm when it works with each TCP. The numerical results based on existing analytic models are also shown.