Summary In this article, an adaptive SR ARQ scheme is analyzed and optimized in a time-varying channel environment with unreliable feedback. Two-state Markov models are assumed for both the forward and the returnchannel.This enables the time behaviour of the entire system to be modelledby a Markov chain. To have an easy-to-implement scheme, we choose the smaller packet size to be half of the bigger one. Thus, there is only oneindependent design parameter related to packet size. The other two parametersare associated with channel sensing. The performance of the adaptive schemeis measured by its average throughput, which is upper-bounded by the throughput of the so-called ideal adaptive scheme. This upper bound is very useful in the optimization of the packet size and other design parameters. From the results, it can be concluded that the selection of the packet size is at least as critical as the optimization of the other parameters. The feedback errors degrade the throughput of the scheme, but their effect on the parameter optimization turns out to be almost negligible.