A performance study of the access channel of a mobile, micro-cellular radio telephone network is presented. The access method used is a combination of the Carrier Sense Multiple Access with Collision Detection (CSMA/CD) and the Busy Tone Multiple Access (BTMA) protocols with modified access and back-off probabilities. Based on an imperfect channel and the likely low connectivity of the mobiles with the base station, two new, semi-Markov renewal models are constructed; the asynchronous time, infinite population model, and the synchronous time, finite population model. The latter uses the mini-slot property of the channel, and it is proven to have a matrix geometric representation. The analysis assumes that capture is due either to slow Rayleigh fading or to power level variations with the distances, and the corresponding capture probabilities are calculated analytically, or by using standard Monte Carlo simulation techniques. With the proposed hybrid protocol applied to both models when the capture is perfect, the performance is similar to the classic CSMA/CD, and thus it is better than the performance obtained by the ALOHA or CSMA algorithms with capture. Nevertheless, the power capture effect does not offer a similar improvement in the maximum achieved throughput, as in the cases of the ALOHA or CSMA contention based protocols. But, if capture due solely to slow fading is present, our hybrid model has considerably better performance than the corresponding CSMA (α = 0.01) and BTMA protocols. However, with the proposed algorithm the channel shows a bi-stable behaviour, unless the offered traffic is extremely low. The analysis allowed us to draw some very useful conclusions for both models developed. Thus, in the case of the synchronous time, finite population model and with respect to the throughput-delay performance, the most important parameters, including the attempt rate, the signal-to-noise ratio and the length of the successful and unsuccessful messages, are considered. For suitable parameter values, and independently of the attempt rate, the channel appears to have an acceptable operational point. In the case of the infinite model, interest has been given in real systems with NCFSK and CPSK modulations, operating in a slow Rayleigh fading environment. Note that if only power capture is present, the same model yields an upper bound on the throughput. The analysis shows that the maximum achievable throughput is obtained for the same value of traffic rate, independently of the packet length. Higher values of the average SNR leads the system to approach the nonfading case.