This paper develops a simple analytic model to characterize the steady-state throughput of a bulk transfer TCP flow, considering both loss rate and round trip time. Unlike previous models, this one captures the behavior of TCP's fast retransmit mechanism and the effect of timeouts on throughput. The authors find that timeouts are more common than fast retransmits in TCP traces, highlighting the importance of including timeout behavior in the model. The model is validated using measurements from 37 TCP connections between 18 hosts across the United States and Europe. The results show that the proposed model provides more accurate predictions of TCP throughput over a wider range of loss rates compared to existing models. The paper also discusses the assumptions underlying the model and suggests future work, including enhancements to account for fast recovery and fast retransmit, and the use of a Markov chain to model the congestion window size.This paper develops a simple analytic model to characterize the steady-state throughput of a bulk transfer TCP flow, considering both loss rate and round trip time. Unlike previous models, this one captures the behavior of TCP's fast retransmit mechanism and the effect of timeouts on throughput. The authors find that timeouts are more common than fast retransmits in TCP traces, highlighting the importance of including timeout behavior in the model. The model is validated using measurements from 37 TCP connections between 18 hosts across the United States and Europe. The results show that the proposed model provides more accurate predictions of TCP throughput over a wider range of loss rates compared to existing models. The paper also discusses the assumptions underlying the model and suggests future work, including enhancements to account for fast recovery and fast retransmit, and the use of a Markov chain to model the congestion window size.