Reliability indicates the internal consistency of a test. In educational studies, reliability is a key feature for a test. Researchers have proposed many traditional reliability estimates, such as coefficient alpha and coefficient omega. However, traditional reliability indices do not deal with the data hierarchy, even though the multilevel structure is prevalent in most of the educational data sets. And then level-specific reliabilities (i.e., the within-group and between-group level) for multilevel data structure have been recently proposed. But the new approach has not considered the influence of missing data, which is very common in educational studies, social and behavioral areas. The aim of this study is to investigate the reliability estimation for multi-level data with missing values. We first reviewed the traditional single level reliabilities and the level-specific reliabilities. And then we proposed a new model, the multilevel confirmatory factor with missing values. This model was illustrated by analyzing a real data set, the 2018 PISA data. We found some differences between the single-level and the level-specific reliability estimates. And then we further investigated the performance of the model by conducting simulation studies under various conditions. We focused on a two-levels single-factor model with six indicators. Six reliability estimates (single-level coefficient alpha [alpha], within-group coefficient alpha [alpha][subscript w], between-group coefficient alpha [alpha][subscript B], single-level coefficient omega [lower case omega], within-group coefficient omega [lower case omega][subscript w], and between-group coefficient omega [lower case omega][subscript B]) were investigated and compared. Various simulations conditions were considered, number of clusters, cluster sizes, intraclass correlation (ICCs), missing data mechanisms, missing data proportions, and missing data techniques. Parameter biases and convergence rates were compared. Results showed that: (1) missing values have negative impacts on the reliability estimation and with the proportion of missing data increased, the percentage bias increased while the convergence rates decreased, (2) listwise deletion method performed worse than the full information maximum likelihood (FIML) method, (3) ICC is importance in the reliability estimation and when ICC was large (say, 0.3 in our simulation), the between-group level reliability estimates outperformed the within-group reliabilities, (4), the coefficient omega performed better than the coefficient alpha if the tau-equivalent assumption is violated. Limitations of the current model and future research were also discussed in the last part. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]