In this paper, we investigate the loss recovery behavior of TCP Reno over wireless links where packets are not always lost due to network congestion. Loss recovery of TCP Reno operates based on two basic algorithms of fast retransmit and fast recovery. We derive the conditions for successful loss recovery of TCP Reno by modeling the window variation during loss recovery accurately. Based on the developed model, the fast retransmit probability of TCP Reno is derived when the pattern of packet losses are random and correlated. Therefore, the performance of TCP Reno can be evaluated from the aspect of the fast retransmit probability. Main findings of our work are 1) loss recovery function of TCP Reno is not effective when multiple packets are lost in a window. 2) Continuity of packet losses as well as packet loss probability can affect the fast retransmit probability. We explain some of these observations in terms of the variation of the average window size and packet loss probability.