To assess inter- and intra-observer errors in identifying the transepicondylar axis and Whiteside's line in a cadaveric model mimicking total knee arthroplasty. Four cadaveric knees with intact soft tissues were used. The knees were exposed anteriorly using the Insall approach, with the patella everted laterally. Three observers (2 surgeons and one trainee) took turns to identify the anatomic landmarks of the transepicondylar axis and Whiteside's line. Each observer repeated the process 20 times. Each identification was photographed and referenced with the true values obtained from the knees after they were stripped of all soft tissue. Inter- and intra-observer errors in the anatomic landmarks were compared. Inter-observer error was significant with both the transepicondylar axis and Whiteside's line (p<0.001, one-way ANOVA). The intra-observer variation was greater for Whiteside's line than the transepicondylar axis (standard deviation, 4.2 vs 2.5 degrees). The maximum potential errors in the transepicondylar axis and Whiteside's line were 13 degrees and 24 degrees, respectively. The accuracy of rotational alignment of the transepicondylar axis and Whiteside's line were operator-dependent, and their intra-operative reproducibility was low.