This study aims to estimate measurement errors in thesis defense evaluations for students in the Mathematics Education Program at Universitas Labuhanbatu. Using a quantitative descriptive approach, the research focuses on breaking down measurement errors in thesis defense evaluations. The population consists of thesis evaluation sheets, and the sample includes responses from three examiners evaluating 10 students. The sampling technique used is *proportionate stratified random sampling*. Each examiner is given an 8-item evaluation sheet measuring 8 aspects. The evaluation data from the three examiners will be analyzed using the *Generalizability Theory Two Facet* with a design of $i \times r \times s$. The analysis reveals that the largest variance, 75.1%, is in the residual aspect, followed by 13.1% in the student-examiner interaction aspect and 11.6% in the student aspect. The ratio of variances-N is highest in the student aspect at 5.12, followed by 0.96 for the student-examiner interaction, 0.23 for the residual, and 0.0045 for the student-item aspect. The initial generalizability coefficient is 0.5096671, which is modified to yield a dependability coefficient of 0.810694. The study highlights the need for consistent and objective evaluation by examiners to reduce measurement errors. It also suggests that increasing the number of examiners and items on the evaluation sheet can further improve reliability. The findings emphasize the importance of addressing the issue of examiner consistency and the need for more objective evaluation methods to ensure fair and accurate thesis defense evaluations.This study aims to estimate measurement errors in thesis defense evaluations for students in the Mathematics Education Program at Universitas Labuhanbatu. Using a quantitative descriptive approach, the research focuses on breaking down measurement errors in thesis defense evaluations. The population consists of thesis evaluation sheets, and the sample includes responses from three examiners evaluating 10 students. The sampling technique used is *proportionate stratified random sampling*. Each examiner is given an 8-item evaluation sheet measuring 8 aspects. The evaluation data from the three examiners will be analyzed using the *Generalizability Theory Two Facet* with a design of $i \times r \times s$. The analysis reveals that the largest variance, 75.1%, is in the residual aspect, followed by 13.1% in the student-examiner interaction aspect and 11.6% in the student aspect. The ratio of variances-N is highest in the student aspect at 5.12, followed by 0.96 for the student-examiner interaction, 0.23 for the residual, and 0.0045 for the student-item aspect. The initial generalizability coefficient is 0.5096671, which is modified to yield a dependability coefficient of 0.810694. The study highlights the need for consistent and objective evaluation by examiners to reduce measurement errors. It also suggests that increasing the number of examiners and items on the evaluation sheet can further improve reliability. The findings emphasize the importance of addressing the issue of examiner consistency and the need for more objective evaluation methods to ensure fair and accurate thesis defense evaluations.