Background Preoperative templating of total knee arthroplasty (TKA) can nowadays be performed three-dimensionally with software solutions using computed tomography (CT) datasets. Currently there is no consensus concerning the axial orientation of TKA components in three-dimensional (3D) planning. Purpose To assess intra-/inter-observer reliability of detection of different bony landmarks in planning axial component alignment using axial CT images and 3D reconstructions. Material and Methods Intra- and inter-observer reliability of determination of four predefined axial femoral and tibial axes was calculated using data from CT scans. Axes determination was performed on the axial slices and on the 3D reconstruction using preoperative planning software. In summary, 61 datasets were analyzed by one medical student (intra-observer reliability) and 15 datasets were analyzed by four different observers independently (inter-observer reliability). Results For the femur, clinical epicondylar axis and posterior condylar axis showed the best reliability with an inter-observer variability of 0.7° and 0.5°, respectively. For the tibia, posterior condylar axis provided best reliability (inter-observer variability: 1.7°). Overall variability was greater for tibial than for femoral axes. Reliability of axis determination was more accurate using axial CT slices rather than 3D reconstructions. Conclusion The femoral clinical epicondylar axis is highly reliable. Landmarks for the tibia are not as easily identifiable as for the femur. The tibial posterior condylar axis presents the axis with highest reliability. Based on these results, clinical epicondylar axis for orientation of the femoral TKA component and posterior condylar axis for the tibial implant, both defined on axial slices can be recommended.