New robotics devices can provide minimally invasive access to deep lesions in the brain through small openings in the skull, which can reduce infection risk and improve recovery times when compared to larger craniotomies. Several groups have introduced semi and fully automatic path planning tools to optimize lesion access while avoiding damage to critical structures along the path. These tools utilize preoperative imaging and intraoperative neuronavigation to plan and execute lesion access. However, several factors can introduce uncertainty into path planning, including segmentation, patient-to-image registration, brain shift during surgery, and mechanical uncertainty in the robotic device during path following. Most systems handle uncertainty by adding a ‘safety margin’ (typically 2–3 mm) around structures to be avoided or around the path. However, safety margins are somewhat arbitrary. We propose using more rigorous models of uncertainty during path planning. We explore two sources of uncertainty in path planning, namely segmentation uncertainty and uncertainty due to brain shift, propose a method for computing brain shift uncertainty, show how to combine these uncertainties into a single risk volume, and provide clinical motivation by presenting our approach in the context of a path planning system for robotic neurosurgery designed for treating mesial temporal lobe epilepsy.