The purpose of this study is to investigate the reliability, validity and vertical equating of the Reading subtest of the Test of Chinese as a Foreign Language. Four levels are included in the reading section, they are Level 2, 3, 4, and 5, respectively. The analysis data was sampled from the formal version of the test administered in 2011 and pretest version in 2012. The results showed that, first, the coefficients of the Kuder-Richardson 20 were closed to or higher than .90. Moreover, large test information is provided to the value of cutoff which is determined an examinee is passed or failed. In other words, low standard error of estimation was obtained for the examinees. Second, the results of factor analysis showed that only one factor was extracted, which could account for above 66% of the variance. In addition, the results of Rasch analysis revealed that more than 87.5% of the items fit the model well. Third, there is a suitable range of difficulties for each level of test. Finally, standard error of estimation about the cutoff values were similar to Level 3 but lower than Level 4 when the items in Level 3 and 4 were split to assemble two tests (i.e., test information on the cutoff values for the even items included in Level 3 and 4, the odd items included in Level 3 and 4, and items in Level 3 and 4). That is these two adjacent levels can be combined to form a composite level of test in the future to reduce the burden for examinees and developers of the test. However, the item difficulty distribution of the composite test should be adjusted.