This study estimates measurement error in the evaluation of thesis defense for students in the Mathematics Education program at Labuhanbatu University. The research uses a quantitative descriptive approach to analyze measurement errors in thesis defense evaluations. The population consists of thesis defense evaluation forms, while the sample includes responses from three evaluators assessing 10 students. A proportional stratified random sampling technique was used. Each evaluator received a 8-item evaluation form measuring 8 aspects. The evaluations from the three evaluators were analyzed using Generalizability Theory Two-Facet with a design of i × r × s. The results showed that the largest variance was in the residual aspect (75.1%), followed by the student-rater interaction (13.1%) and the student aspect (11.6%). The largest ratio of variance-N was in the student aspect (5.12), followed by the student-rater interaction (0.96), residual (0.23), and student-item (0.0045). The initial generalizability coefficient was 0.5096671, which was modified to a dependability coefficient of 0.810694. The study found that the largest variance in the G-study was in the residual aspect, indicating that there were factors not considered in the analysis affecting the thesis defense evaluation. The study also found that the student aspect had a relatively small variance (11%) but had the highest D-study value. The D-study analysis showed that increasing the number of evaluators and items reduced error. The dependability coefficient increased to 0.810694, indicating that the measurement error in the thesis defense evaluation could be avoided by increasing the number of evaluators and items on the evaluation form.This study estimates measurement error in the evaluation of thesis defense for students in the Mathematics Education program at Labuhanbatu University. The research uses a quantitative descriptive approach to analyze measurement errors in thesis defense evaluations. The population consists of thesis defense evaluation forms, while the sample includes responses from three evaluators assessing 10 students. A proportional stratified random sampling technique was used. Each evaluator received a 8-item evaluation form measuring 8 aspects. The evaluations from the three evaluators were analyzed using Generalizability Theory Two-Facet with a design of i × r × s. The results showed that the largest variance was in the residual aspect (75.1%), followed by the student-rater interaction (13.1%) and the student aspect (11.6%). The largest ratio of variance-N was in the student aspect (5.12), followed by the student-rater interaction (0.96), residual (0.23), and student-item (0.0045). The initial generalizability coefficient was 0.5096671, which was modified to a dependability coefficient of 0.810694. The study found that the largest variance in the G-study was in the residual aspect, indicating that there were factors not considered in the analysis affecting the thesis defense evaluation. The study also found that the student aspect had a relatively small variance (11%) but had the highest D-study value. The D-study analysis showed that increasing the number of evaluators and items reduced error. The dependability coefficient increased to 0.810694, indicating that the measurement error in the thesis defense evaluation could be avoided by increasing the number of evaluators and items on the evaluation form.