In vivo measurements of muscle architecture provide insight into inter-individual differences in muscle function and could be used to personalise musculoskeletal models. When muscle architecture is measured from ultrasound images, as is frequently done, it is assumed that fascicles are oriented in the image plane and, for some measurements, that the image plane is perpendicular to the aponeurosis at the intersection of fascicle and aponeurosis. This study presents an in vivo validation of these assumptions by comparing ultrasound image plane orientation to three-dimensional reconstructions of muscle fascicles and aponeuroses obtained with diffusion tensor imaging (DTI) and high-resolution anatomical MRI scans. It was found that muscle fascicles were oriented on average at 5.5±4.1° to the ultrasound image plane. On average, ultrasound yielded similar measurements of fascicle lengths to DTI (difference <3 mm), suggesting that the measurements were unbiased. The absolute difference in length between any pair of measurements made with ultrasound and DTI was substantial (10 mm or 20% of the mean), indicating that the measurements were imprecise. Pennation angles measured with ultrasound were significantly smaller than those measured with DTI (mean difference 6°). This difference was apparent only at the superficial insertion of the muscle fascicles so it was probably due to pressure on the skin applied by the ultrasound probes. It is concluded that ultrasound measurements of deep pennation angles and fascicle lengths in the medial gastrocnemius are unbiased but have a low precision and that superficial pennation angles are underestimated by approximately 10°. The low precision limits the use of ultrasound to personalise fascicle length in musculoskeletal models.