Augmented reality (AR) improves science education by facilitating teaching and experiments in schools and universities. Studies show positive effects, like increased motivation and improved concept connections, but there is a lack of consistency in the implementation and investigation of AR in science experiments. This review examines AR usage, criteria, design parameters for the development of AR applications and the validation methods, taking into account the PRISMA guidelines. A Web of Science database search using “Publish or Perish” software (version 23.4.0) identified 247 potentially relevant articles from 2000 to March 2024 in international, peer-reviewed journals. After removing duplicates, inaccessible abstracts, and applying inclusion and exclusion criteria, 40 studies were selected for in-depth analysis. Physics had the most AR applications, primarily for visualizing invisible properties. Most studies used quantitative or qualitative methods, only a few used both or did not conduct empirical research. Research questions varied, but common drawbacks included small sample sizes and low use of AR design parameters such as interactivity, adaptivity, realistic representation and use of game elements. This review identifies opportunities for improvement in the implementation and investigation of AR in science education experiments and emphasizes consistent and rigorous approaches to fully exploit the benefits of AR in science education.