In the learning process, most exams to assess learning achievement have been carried out by providing questions in the form of short answers or essay questions. The variety of answers given by students makes a teacher have to focus on reading them. This process of assigning grades is difficult to guarantee quality if done manually. Moreover, each class is mastered by a different teacher, which can cause inequality in the grades obtained by students due to the influence of differences in teacher experience. Therefore the automated answer assessment research was developed. The automatic short answer assessment is designed to automatically assess and evaluate students' answers based on a trained set of answer documents. The automated grading system uses the cosine similarity method to determine the degree of similarity of a student's answer to the teacher's answer. While the word weighting used is the Term Frequency-Inverse Document Frequency (TF-IDF) method. The data used is a question totaling 5 questions with each question answered by 30 students, while the students' answers are assessed by experts to determine the real value. This study was evaluated by Mean Absolute Error (MAE) with the resulting value of 0.22.