The PISA 2006 science assessment is composed of open response, multiple-choice, and constructed multiple choice items. The current study introduced the random item response models to investigate the item format effects on item difficulties, and these models include the linear logistic test model with raThe PISA 2006 science assessment is composed of open response, multiple-choice, and constructed multiple choice items. The current study introduced the random item response models to investigate the item format effects on item difficulties, and these models include the linear logistic test model with random item effects (i.e., the LLTM-R) and the hierarchical item response model (i.e., the hierarchical IRM). In this study these models were applied to the PISA 2006 science data set to explore the relationship between items' format and their difficulties. The empirical analysis results in the PISA 2006 science assessment first find that the LLTM-R and the hierachical IRM provides equivalent item difficulty estimates compared with those from the Rasch model and the LLTM, and also clearly show that the item difficulties are substantially affected by item formats. This result implies that item difficulties may be different to each other depending on the item format although they deal with the same content.ndom item effects (i.e., the LLTM-R) and the hierarchical item response model (i.e., the hierarchical IRM). In this study these models were applied to the PISA 2006 science data set to explore the relationship between items' format and their difficulties. The empirical analysis results in the PISA 2006 science assessment first find that the LLTM-R and the hierachical IRM provides equivalent item difficulty estimates compared with those from the Rasch model and the LLTM, and also clearly show that the item difficulties are substantially affected by item formats. This result implies that item difficulties may be different to each other depending on the item format although they deal with the same content.