Cognitive activation strategies require students to think more deeply in order to reach the answer. These strategies emphasize the ways that the students utilize while responding to a question. The process of cognitive activation exerts a great impact on several student characteristics which are closely tied to concepts such as achievement motivation, self-efficacy and self-concept,  and helps decrease the anxiety that the students have.  Since measuring these strategies more accurately is of utmost importance, it is fair to say that more qualified items should be selected. At this point, the calibration of the items representing these strategies comes into play. In this regard, for the cognitive activation strategies scale is a 4-point Likert scale with 9 items, the aim of this study was to compare different IRT models, namely Graded Response Models (GRM) and Partial credit model (PCM) in terms of theta estimation, information and reliability and to determine which of the models at hand fit the data better. To this end, the data used in the present study were obtained from 2012 Turkish administration of PISA (N=3076). Analysis were performed by using MULTILOG 7.03 software. The reliability coefficient of GRM was found to be higher than that of PCM. It was also found that the GRM had lower -2loglikelihood value and the difference between -2loglikelihood values of these two models was significant (p<.05). All these indicated that GRM fitted the data better. The findings of this study raised a number of important implications, and pointed to some possible directions for future research.