The purpose of this study is to explore the authenticity of referential tasks in speaking tests of English. All speaking does not always have the attribute of referential communication; it is necessary to critically review the referential tasks because they dominantly appear in the recently used computer-based speaking tests of English. This study collected 118 question items from the official test preparation materials and homepage in TOEIC Speaking, OPIc, and NEAT Speaking. Among them, the question items containing features of referential communication tasks were 80 in total. The test items were categorized into task classifications by Yule (1997), and all the categorized tasks were analyzed using the three types of referential communication features (substantive knowledge, enabling skill, procedural knowledge) framed by Sonnenschein and Whitehurst (1984). It was found that test rubrics did not allow speakers of referential tasks to fully utilize their own substantive knowledge. Enabling skills were also restricted to the use of elaborated vocabulary. During limited time of speaking, or in short answer-type responses elicited, speakers may make it difficult to picture the referential goals. Procedural knowledge was scarcely performed because any role of collaborative listeners is not available. Alternative rubrics of test tasks are suggested to attract the attention of future task designers and researchers.