Measurement models using maximum likelihood estimation methods are commonly used to estimate latent traits of person from their responses in a test. However, whether the estimated obtained from such estimation is heavily affected by whether the person responses are “accurate”. “Accuracy” of person measures may be affected by a number of personal characteristics that may not well anticipated from the researchers’ point of view, such as guessing behaviours and carelessness, making the responses contaminated. As such, the estimation of latent trait from these responses will become contaminated too. In other words, the claimed model to recover the latent trait will not be robust. This article, therefore, aims to propose a method that is robust to estimate latent trait from IRT models. By using weighted functions, that is bisquare and Huber, and adding arbitrary constants. Person parameters were compared in the simulation studies to test the level of robustness of the proposed methods. The applicability of the proposed methods is easily understood: some measurement of person latent traits in testing situations where person rankings are relevant and important. However, it is expected that the model might not be easily practicable for high-stake testing situations since stakeholders in such situation might probably question for the rationale for such measurement and ranking methods when the stake of the testing results is high. Nevertheless, the article has demonstrated how person estimates might be affected by the mentioned sources of disturbances and how this issue could be tackled. Despite the potential benefits provided by the new adaptation method, it is worth asking about the degree of robustness of such methods and the testing of it so that the effectiveness of using these methods over traditional methods could be assessed and reported.