This paper presents a solution to the problem of self-localisation of mobile nodes in wireless sensor networks with unknown measurement noise characteristics. A Gibbs sampling algorithm estimates the unknown noise parameters followed by localisation of mobile nodes with a multiple model auxiliary particle filter (MM AUX-PF). The performance of the Gibbs sampler and MM AUX-PF is investigated in terms of accuracy and computational complexity using simulated and real received signal strength measurements. We show that the method provides accurate estimation results with complexity suitable for real-time applications.