Abstract: For the purpose of evaluating the efficacy of description dosage in a clinical trial in which compliance is observed, a new parameter estimation method is proposed for a structure linear model based on the combination of least square and pseudo-likelihood, and under general regular conditions, the consistency and asymptotic normality of the estimators are verified. This method not only is simple and has good properties for large samples, but also has been shown by a group of Monte Carlo simulations having smaller mean square errors for moderate and small sample sizes.