The modeling of TCP transfer latency has received significant attention in the last decade. Several models have been proposed for TCP performance under various conditions. All the available,models predict TCP performance for-a single link. Furthermore, all models relate timeouts to packet drops either due to congestion or due to transmission errors. However, TCP connections may be running over a multilink connection that aggregates the bandwidth of multiple links into a single logical pipe using the multilink point-to-point protocol (MLPPP). In such aggregate links, packet drops occur if any of the individual links experience a call drop. None of the available models account for call drops as a possible source of performance degradation. In this letter, we study the call drop phenomenon under MLPPP and incorporate our results into a method that predicts TCP latency for a long transfer. The performance model is experimentally evaluated by running TCP over MLPPP over multiple Iridium satellite links.