The present study tested the accuracy of an image-free navigation system used for total hip arthroplasty (THA). Two parallel, prospective studies were performed: one on real patients and the other on pelvic phantoms. We used a comparison between the intra-operative cup orientation, as displayed by the navigation system, and the post-operative cup position, as measured on CT data. The mean intrinsic overall error (+/- standard deviation) found in the phantom study was 2.6 +/- 1.1 degrees (range: 1.5-4.4 degrees ) for cup abduction, and 0.9 +/- 0.7 degrees (range: 0-2.5 degrees ) for cup anteversion. The system was less accurate in the clinical operative setting. The evaluation model was able to identify, and more importantly quantify, the clinically induced error. Ameliorating this would improve the clinical accuracy of the system.