There are two different outliers, i.e outlier in observations and outlier in models. The existing outlier detection method in models is using common Likelihood method. The limitation of this method is the optimal value produced might be not the real optimal values. This research yields a method for outlier detection in multivariate linear models with Likelihood Displacement Statistic-Lagrange method (LDL method). This method uses multiplier Lagrange with constraint the confidence interval of parameter’s vector. This parameter’s vector is obtained from the data set which is outlier free. This parameter estimation process uses numerical method with Karush-Kuhn Tucker condition in nonlinear programming. This method compares between LDL value and the table F value that follows the distribution of F value to indentify the outlier in models.