A set of speakers in an arbitrary 3D array can be used to create virtual sound source images to a large audience. The problem of maximising directionality and minimising panning artefacts is described in a mathematical context. An algorithm is proposed to maximise the directionality across the listening region. This solution is analysed and compared to fixed order sound-field techniques. Additions to the algorithm are provided for reducing panning artefacts when animating sound sources. The proposed algorithms have been implemented and auditioned with favourable results.