A Cook-type distance is considered for investigating the influence of observations in constrained regression models. Its exact sampling distribution is derived, which is used for judging whether each observation is influential or not. A numerical example is provided for illustration.