Model-based predictive approach to finite population estimation provides insights into underlying mechanisms of social or economic phenomena. This feature is drawing the favor of specialists in many scientific fields to this method. Its use does not contradict, in most cases, the design-based estimators while taking full advantage of the theory and empirical results of generalized liner models. However, erroneous outlying observations resulting from the survey data collection may cause greater damage and affect the estimation by the model-based methods in a different way. This study proposes an outlier-resistant model-based estimator by bounding the influence function of possible erroneous outliers on estimated model parameters. It examines and compares the effects of outlying/extreme observations on finite population estimators by alternative estimators assuming the super population model is true.