Background: For over a decade, computer-assisted orthopaedic surgery for total knee arthroplasty has been accepted as ensuring accurate implant alignment in the coronal plane. Hypothesis: We hypothesised that lack of accuracy in skeletal landmark identification during the acquisition phase and/or measurement variability of the infrared optical system may limit the validity of the numerical information used to guide the surgical procedure. Methods: We built a geometric model of a navigation system, with no preoperative image acquisition, to simulate the stages of the acquisition process. Random positions of each optical reflector center and anatomic acquisition point were generated within a sphere of predefined diameter. Based on the virtual geometric model and navigation process, we obtained 30,000 simulations using the Monte Carlo statistical method then computed the variability of the anatomic reference frames used to guide the bone cuts. Rotational variability (a, b, g) of the femoral and tibial landmarks reflected implant positioning errors in flexion-extension, valgus-varus, and rotation, respectively. Results: Taking into account the uncertainties pertaining to the 3D infrared optical measurement system and to anatomic point acquisition, the femoral and tibial landmarks exhibited maximal alpha (flexion-extension), beta (valgus-varus), and gamma (axial rotation) errors of 1.65 • (0.9 •); 1.51 • (0,98 •), and 2.37 • (3.84 •), respectively. Variability of the infrared optical measurement system had no significant influence on femoro-tibial alignment angles. Conclusion: The results of a Monte Carlo simulation indicate a certain level of vulnerability of navigation systems for guiding position in rotation, contrasting with robustness for guiding sagittal and coronal alignments. Level of evidence: Level IV.