This study aimed to determine the influence of Assurance, Relevance, Interest, Assessment, Satisfaction (ARIAS) learning models on students' mathematical writing skills reviewed from self-efficacy. The type of research conducted in this study is a quasi-experimental research with a post-test nonequivalent control group design. The population in this study was students majoring in mathematics education at Bina Bangsa University, then two classes are selected at random as research subjects. The data was obtained from mathematical writing ability tests and self-efficacy questionnaires, then processed with ANOVA (Analysis of Variance) tests. The results showed significant differences in mathematical writing skills between students who obtained ARIAS learning models with students who received regular learning, and there were significant differences in students' mathematical writing skills reviewed on their self-efficacy. The implication is known that students with high self-confidence have better writing skills than students whose confidence is low.