The performance of a collision frequency shift keying system (CFSK) multiple access system is investigated. In CFSK two or more users can simultaneously transmit the same frequency resulting in a collision. A new iterative consensus decoder, the maximum a posteriori consensus decoder, is formulated and the performance investigated in the presence of noise. In this decoder, each user uses a maximum a posteriori decoding algorithm to calculate the probability that they transmitted each of the frequencies. The resulting probability distributions are then shared among the users and refined by iteration. This decoder is based on the derivation of a metric for the noisy CFSK channel. Simulation results demonstrate the efficacy of the new decoder.