Objective Early prediction of stem version aids in optimization of combined version during total hip arthroplasty (THA). This study aimed to analyze the discrepancy between stem version and native femoral version measured by different methods, and to explore which method can better predict the stem version. Methods We retrospectively reviewed 26 patients (39 hips) treated with robot‐assisted THA in our hospital between September 2019 and December 2019. A straight, single‐wedge, cementless stem (Accolade II) was used in all cases. Preoperative femoral version was measured at three levels on computerized tomography (CT) scan from the top to the middle level of femoral neck (Level 1 to Level 3). During THA, the version on cutting surface was measured prior to femoral broaching based on two reference lines: mid‐cortical line and T line (trochanteric fossa to the middle of medial cortex). After femoral broaching, stem version was measured based on the femoral neck trial using Mako system (Stryker). In the statistical analysis, the difference and absolute discrepancy between stem version and femoral version measured with various methods were examined using paired t‐test, and the relationship between stem version and various femoral versions were examined using correlation analysis. Results Mean femoral neck version (Level 1) was 9.5° ± 2.6° (range, −16.8°‐42.5°), while mean stem version measured by Mako system was 19.9° ± 2.0° (range, −8.0°‐49.0°). Femoral version measured with each method showed a moderate correlation with stem version (p