Speech signals employed in clinical and research contexts are thought to be realistic or representative to the extent that they consist of phonetic, lexical, and morphosyntactic content. These characteristics may be assumed to ensure that speech perception tests are representative of the demands of real-world speech perception and understanding, and therefore that these tests are predictive of real-world speech understanding. However, there are numerous reports of discrepancies between the results of speech perception tests and real-world speech understanding and hearing device benefit. To make optimal use of existing speech tests, and to design more predictive speech tests in the future, it is important to consider how clinical and research speech tests do and do not represent the process and demands of everyday speech understanding. Speech and language examples from a single talker engaged in conversation with a hearing-impaired interlocutor in a range of realistic acoustic environments will be contrasted with widely used recordings of BKB sentences produced by the same talker. The absence of much natural variation in standard speech test materials, as well as a failure to quantify perception of information which cannot be captured orthographically, will be discussed as limitations to the generalizability of standard speech test results.