This study was intended to explore the improvement of scores that measure students' scientific explanation of light through phenomenon-based learning. Students conduct an investigation and make an explanation about the phenomenon being observed. The study employed mixed methods with an embedded experimental model. The subject consisted of 28 students of eighth graders in junior high school 3 of Malang, Indonesia. Test, interview, and observation collected the data. Pre-test and post-test scores were analyzed with a paired sample t-test, N-gain score, and d-effect size. The data of pre-test and N-gain score were analyzed with product moment Pearson correlation test. The quality of scientific explanation was collected based on the component that consists of claim, evidence, and reasoning. The result showed that the average of students' scientific explanation scores significantly improved from 1,39 in pre-test into 5,68 in post-test with adeffect size of about 3,6 which was categorized as a strong influence and an average N-gain of 0,40 which was categorized as a medium. Students' scientific explanation shifted significantly from invalid category to fully supported category. The benefits of learning for the improvement of scientific explanation have been obtained by the students equally.