Abstract Asymmetric resection of the patella during total knee arthroplasty (TKA) correlates with anterior knee pain, bony impingement and patellar maltracking. Despite this, there is no consensus regarding the desired landmarks; the cut is often done freehand; and there has been no quantitative comparison of proposed resection planes. The objectives of this study were to: determine the intra- and inter-surgeon repeatability of two radiographic resection definitions (medial–divot, MD, and medial–lateral extents, MLE); calculate two additional definitions from the radiographic patellar circumferences (parallel to the anterior surface, ANT, and perpendicular to the anteroposterior tangent points, PERP); compare the clinical resection line to the previous four definitions before and after introducing the MD method clinically; and identify distinguishing features of patellae with better vs. worse resection angles. We hypothesized that the MD method would improve repeatability both radiographically and clinically, that the different radiographic definitions would produce comparable angles, and that we could identify distinguishing features. For the radiographic study, three surgeons drew lines on 40 preoperative X-rays plus 9 interspersed repetitions of 3 of these X-rays. For the clinical study, we compared the patellar resection angle for 20 patients immediately before and after implementing the new method. Given that the clinical goal is to have equal distances from the resection surface to the anterior surface, we compared all results to the ANT definition as the theoretically ideal definition. Confirming the first hypothesis, intra-surgeon repeatability (10 repetitions of 3 X-rays) and inter-surgeon repeatability (3 surgeons×40 X-rays) were both significantly better using the new MD method compared to the MLE method ( p<0.001). Contrary to the second hypothesis, clinical use of the MD method did not improve resection symmetry. Contrary to the third hypothesis, the PERP definition was significantly different from the other three definitions. In agreement with the fourth hypothesis, female patellae and more deformed patella had significantly greater asymmetry ( p<0.001). Given the inherent variability shown in drawing the ‘patellar horizon’, we encourage researchers to draw the line several times and average the results when comparing tilt or the resection angle to this horizon. Based on the distinguishing characteristics of asymmetrically resurfaced patellae in our series, we recommend that clinicians be particularly careful when resecting laterally deformed patellae and the patellae of female patients.