This paper addresses the auto-tuning of a feedforward controller for a linear stage. In our approach, particle swarm optimization (PSO), which is one of meta-heuristic algorithms, is employed. In PSO-based controller tuning, the initial placement and the number of particles are determined a priori. However, the initial placement of particles of the PSO was decided randomly or was not sufficiently discussed. Moreover, although the large number of particles is effective in terms of tuning accuracy, it causes the increase of tuning time. For these reasons, we focus on the setting of the initial placement and the number of particles in order to implement the PSO algorithm appropriate for the controller tuning of the linear stage. To shorten tuning time and improve tuning accuracy, firstly, the initial placement is determined by using the information on the nominal model of the linear stage. Then, since some parameters of the linear stage are known, the dimensionality of particles is reduced. Finally, in consideration of precision positioning, particles are divided into some groups. Furthermore, the number of particles is decided on the basis of the tuning time and the maximum number of iteration. The performance of our initial setting methods is evaluated by simulation and experiment.