Diagnostic test plays an important role in personalized eLearning by providing information about cognitive levels of students? learning states. While many diagnosis algorithms have been proposed, most of them lack a solid theory base. On the other hand, item response theory (IRT) is a widely-accepted test theory and has been shown very effective in estimating a learner?s latent ability. However, it did not tell much about conceptual cognitive states. This paper proposes an extension of IRT model for diagnostic test by extending it into a componential IRT model. This diagnostic test feature has been added into a standard-conformant eLearning system, called IDEAL, where a model-based diagnosis and remediation architecture is implemented. Preliminary results show that the proposed approach is effective.