The limiting throughput efficiency of the classic go-back-N automatic repeat request (ARQ) scheme for burst-error channels described by E. N. Gilbert's (1960) two-state Markov model is evaluated based on regenerative theorems. In Gilbert's model, the channel has two possible states, a quiet state and a noisy state. Let p and q denote respectively the probabilities of transition from the quiet state or the noisy state to itself. Numerical results reveal that the classic go-back-N ARQ scheme performs better under Gilbert's model than an independent error model if and only if p + q ≥ 1. The technique presented can be applied to evaluate the performance of a wide variety of ARQ schemes for burst-error channels.