It is a common observation that toll plazas on highways act as bottlenecks as the vehicles need to stop or at least slow down to pay the toll. In order to improve the design of toll plazas one needs to understand and analyse the queuing process at these facilities. One of the key inputs for such an analysis is how drivers choose toll-lane. In this paper a random-coefficients based multinomial logit model has been proposed to model the toll-lane choice behaviour of the drivers at a toll plaza. The model has been calibrated using toll-lane choice data collected in India. The numerical simulation technique used for estimating the mixed logit model is Halton sequence based Quasi-Monte Carlo method and the optimisation algorithm used to search for the parameter estimates which maximise simulated log likelihood is trust region method.