Statistical literacy is essential to people of modern era. However, research has shown that students often hold misconceptions when they study statistics. Although two-tier diagnostic instruments have been developed for diagnosing and exploring students’ misconceptions in other fields of science, few are availble in statistics education. Therefore, the main purpose of the current study was to develop a two tier diagnostic instrument: Two-tier Diagnostic Instrument about Correlation Misconceptions (TDICM). This instrument was adopted to analyze the responses of 982 12th-graders who had taken statistics courses to “correlation”, and to explore the types, universality, and possible causes of such misconceptions in correlation. The research results showed that: (a) TDICM has satisfactory levels of difficulty, internal consistency reliability, and discrimination; (b) most participants do not have good comprehension of “correlation” concepts. Although some were able to answer the questions correctly in the first tier of TDICM, they failed to do so in the second tier due to lack of comprehensive understanding of correlation; (c) students’ responses in the two-tier model indicated that there are 14 correlation misconceptions universally held by the participants; (d) seven of the 14 misconceptions has been discussed in the literature, and others were identified in this study. Furthermore, the possible causes of these misconceptions and the implications in education were discussed. Finally, this study provided some suggestions to the application of two-tier diagnostic instrument and the correlation misconceptions in statistics education.