This study is to investigate the accuracy of ability parameter estimates based on the DINA model using MCMC and MMLE/EM algorithms through several simulation conditions, focusing on the two kinds of results such as mastery probability and mastery state. The former can be performed by the WinBUGS program, and the latter can be utilized using the Ox program which provides information only on mastery state of each attribute. In a simulation study to evaluate the accuracy of ability recovery, two factors such as the test lengths and the correlations among cognitive attributes were considered together. The results show that if one has to choose between mastery state and mastery probability to evaluate examinee’s cognitive ability, presenting mastery state rather than mastery probability can be safer choice. Comparing the performances of MCMC and MMLE/EM algorithms in terms of classification accuracy, the latter appears to provide consistently better estimates than the former. Moreover, the result suggests not only that the longer test length is given, the larger classification accuracy tends to be obtained, but also that accuracy classification results at the whole-pattern level increases when the correlations between attribute are higher. Such relationship between the correlations and classification accuracy, however, was not found when the recovery is evaluated in terms of the mastery on each individual attribute.