The performance of an adaptive token passing bus protocol is analysed by modelling the behaviour of each node as an M/GI/1 queue and deriving a set of fixed point equations. The technique is shown to be applicable to the analysis of a number of variants of the token passing bus protocol which arise in communication networks and computer integrated manufacturing systems. The validity of the independence assumption underlying this approach is investigated by first examining the autocorrelation function of the simulated time series of token passing cycles. This indicates that cycles are not highly correlated except at high traffic loads. Mean message transmission times predicted by the models are compared with the corresponding results of simulation experiments and the performance of the adaptive protocol is compared quantitatively with that of the standard protocol.