The purpose of this study was to develop a computer-based measure of elementary students' science talk and to report students' benchmarks. The development procedure had three steps: defining the framework of the test, collecting and identifying key reference sets of science talk, and developing and verifying the science talk instrument. The Science Talk Test (STT), consisting of semi-structured items, was developed for a framework of exemplars from inquiry classrooms representing three inquiry components (i.e. identifying questions, designing methods and presenting evidence, and drawing evidence-based conclusions) for two types of science talk (i.e. factually oriented and evaluative discourses). Grades 4 and 6 students (N = 281) were selected from three levels of urbanization representative of Taiwan to verify the STT and to establish performance benchmarks. The STT's reliability, validity, and item parameters were found to be reasonable. Students performed better in the factually oriented talk than in the evaluative talk. Designing methods and presenting evidence was more difficult than the other two types of inquiry talk. Students' science talk ability was predictable with prior scientific knowledge and language ability measures. Practical use of the STT, explicit instruction, and future research are recommended.