In this paper, several modifications of an efficient ARQ scheme proposed by Weldon are studied. Unlike Weldon's scheme in which all erroneous data packets are discarded, the present schemes make use of copies of the data packet which may contain errors. A number of channel models are considered, namely a binary symmetric channel, a nonfading, and a Rayleigh fading channel with additive white Gaussian noise. In most cases, it is found that the throughput can be substantially increased. Under poor channel conditions, the use of forward error correction can lead to further improvement. A type II ARQ scheme which does not suffer the throughput degradation under good channel conditions due to overhead parity bits associated with conventional forward error correction is also analyzed.