Large scale assessment routinely release part of the sample items to communicate the assessment theme. Proportion correct of each release item is also included in the release documentation. To translate the statistic information into teaching practice adjustment, teachers usually need some professional supports. In this study, an analysis framework on item cognitive complexity is proposed and implemented. The 2005 and 2006 on-line tests of the Taiwan Assessment of Student Achievement in Mathematics (TASA-MAT) for the 6th graders were used for the preliminary analysis. A 6 cognitive components coding schema was developed to predict the item difficulty parameters. The results suggest that the framework proposed can predict around 27% of the difficulty variance across five contents. Within some sub-contents, one to two cognitive components can account up for 50% of the difficulty variance. The implications of these results for math teachers are discussed.