Augmented reality (AR) is becoming essential in several surgical specialities. Fusing the patient-specific preoperative planning information, typically 3D models extracted from CT scans or MRI, on the endoscopic images, it allows the surgeons to have thorough and detailed knowledge of the anatomical structure of the surgical target intra-operatively. An AR system’s first step consists of an initial rigid registration of the 3D models on the surgical image. Most of the approaches are manual or semi-automatic. While the latter exploits physical landmarks on the tissue or 3D surface features, the manual approach requires Human-Computer Interfaces (HCI) to manipulate the models. The choice of the HCI is fundamental in order to develop an easy-to-use application with a slight learning curve and supporting the surgeon. In fact, the way the manual registration is performed is strictly connected to the users’ physical and mental stress, influencing their final performances. This paper presents a human-centred usability study that aims to evaluate the most appropriate HCI for 3D model manipulation in AR surgical applications. A software interface was developed to control the 3D model manipulation using SpaceMouse ® and UltraLeap ® . Subjects were asked to perform the rigid manual registration of the 3D model of a kidney on the top of an image, showing the real kidney. Quantitative and qualitative evaluation was performed to select the HCI that minimizes the cognitive and physical load of the operators and maximize their performance.