Musculoskeletal models rely heavily on the use of an anatomical dataset and clearly defined assumptions to accurately model the subject being studied. Therefore, it is important to understand the limitations of using musculoskeletal models to study individuals. This paper describes a method of measuring in vivo gracilis muscle-tendon unit length and presents a comparison of experimental data versus predictions from four musculoskeletal models in OpenSim. The largest errors occurred when the knee was fully extended. At this position, the absolute average muscle-tendon unit length error was 7% and the absolute average fiber length error was between 15% and 32%. However, the variability of these errors was significant. Manual linear scaling based on an anthropometric database did not capture the variability observed in subjects. The fiber length errors observed are predicted to have a significant impact on muscle force production that may not represent true subject specific force-length relationship of the gracilis.