A novel constraint using shape parameter distribution into the AAM fitting method is proposed. Active appearance models (AAMs) are some of the most popular facial models. AAM-based face tracking delivers accurate alignment results. However, non-face-like shapes can also be estimated by AAMs, unlike by the conventional AAM fitting method, which only minimizes the matching error of the image. This is one of the causes for face tracking performance degradation in AAMs. A constraint using the shape parameter distribution is added in order to solve this problem.