In this paper, the impact of user relaying on the behavior of a relay node, which acts as the source node at the same time, is analyzed in a wireless relay network at the packet level. The analysis process models the behavior of the relay node as a queueing system and represents the service for its own packet transmission as an M/G/1-type Markov chain. By considering the fact that the maximum number of packet arrivals is ordinarily limited in a practical system, the M/G/1-type Markov chain is further reformatted into a quasi-birth-death (QBD) process through re-blocking so as to simplify the analysis and obtain the associated performance, such as average packet transmission delay. As an application of the results arising from the analysis, a new relay node selection scheme, based on a utility function approach that jointly considers the channel and the queue conditions at the relay node, is proposed. Numerical results show that the proposed analysis model is quite accurate and the prop...