The hypothesis of this study was that length, torsion, and axis of a leg phantom can be measured accurately and reproducibly by magnetic resonance imaging (MRI) and that this can be applied to patients with leg deformities. Two phantoms and 30 patients (genu varum, n = 15; genu valgum, n = 15) were investigated using an optimized MRI technique. Reference measurements were performed with a micrometer screw and a goniometer. Patient leg length and axis were compared with long radiographs in bipedal stance. Intra- and interobserver reproducibility and accuracy were calculated using the mean absolute difference (MAD) and the 95% confidence interval. In patients, comparisons were done using a paired Student's t-test. MAD, intraobserver MAD, and interobserver MAD were 0.03, 0.03, 0.04 mm (length); 0.98, 1.2, 0.98 degrees (torsion); and 0.18, 0.23, 0.22 degrees (axis), respectively. In patients, leg length was underestimated by MRI (-2.4 +/- 0.7%; 1.9 +/- 0.7 cm; P < 0.001). The hip-knee-ankle angle (HKA) did not show significant differences in varus knees (-0.5 +/- 1.0 degrees ; P > 0.05), while it was significantly underestimated in valgus knees (-3.6 +/- 2.8 degrees ; P < 0.05). The phantom study revealed that leg length, torsion, and axis can be measured accurately and reproducibly by MRI. Although underestimation of leg length and HKA in valgus knees occurred, this optimized MRI technique can be applied to patients with leg deformities.