Augmented reality (AR) is currently considered as having potential for pedagogical applications. However, in science education, research regarding AR-aided learning is in its infancy. To understand how AR could help science learning, this review paper firstly has identified two major approaches of utilizing AR technology in science education, which are named as "image-based AR" and "location-based AR." These approaches may result in different affordances for science learning. It is then found that students' spatial ability, practical skills, and conceptual understanding are often afforded by image-based AR and location-based AR usually supports inquiry-based scientific activities. After examining what has been done in science learning with AR supports, several suggestions for future research are proposed. For example, more research is required to explore learning experience (e.g., motivation or cognitive load) and learner characteristics (e.g., spatial ability or perceived presence) involved in AR. Mixed methods of investigating learning process (e.g., a content analysis and a sequential analysis) and in-depth examination of user experience beyond usability (e.g., affective variables of esthetic pleasure or emotional fulfillment) should be considered. Combining image-based and location-based AR technology may bring new possibility for supporting science learning. Theories including "mental models," "spatial cognition," "situated cognition," and "social constructivist learning" are suggested for the profitable uses of future AR research in science education.