The aim of this study was to determine the reliability of bone landmarks for restoring the joint line in revision knee arthroplasty. The relationship of the femoral epicondyles, the tibial tubercle (TT) and the fibular head (FH) to the joint line was measured on 200 magnetic resonance imaging (MRI, 100 females, 100 males), including assessment on intraobserver and interobserver reliability. MRI scans demonstrating chondral lesions and osteoarthritis were excluded, as were patients with immature skeletons or a history of previous knee surgery. Sequences in sagittal, coronal and axial planes were used as well as cross-referencing with the same computer software. In order to account for size differences between patients, each bony landmark measurement was converted to a ratio relative to the femoral or/and tibial width. We found a transepicondylar axis equal to 3.11 degrees (+/-1.9). The average distance from the epicondyles to the joint line was respectively 23 mm on the lateral side and 28 mm on the medial side. However there was a variation of distance from the epicondyles of the joint line up to 11 mm and a significative difference was found between male and female. We determined the distances from the tip of the FH and from the TT to the joint line. The joint line-FH distance averaged 14 mm (range 4.1-22.13) with no gender difference. The joint line-TT distance was averaged 22 mm (range 10.61-32.09). We determined an epicondylar ratio (distance from the lateral epicondyle to the joint line related to the femoral width). We found this ratio averaged 28% with no gender difference (P = 0.09). There is a large variation of bony landmarks depending on the size of the individual. Considering this findings, the FH is not a reliable guide for the joint line in revision surgery. Previous studies have measured the absolute values from various landmarks to the joint line. This study provides a significant advantage, in that the level of the joint line can be determined for each individual by using a ratio to account for gender and size differences.