Clinical assessments of speech-perception difficulties involve speech-in-noise tests in which individuals recognize words (or sentences) at varying signal-to-noise ratios (SNRs). Words are presented in lists that have been balanced based on their phonetic properties or on their average intelligibility in quiet from norming studies. Such list-wise balancing presumably allows for generalizations to be drawn about the difficulty of an SNR irrespective of word-level difficulty. However, differences in the lexical properties of items (e.g., lexical frequency) may yield substantial variability in how hard individuals work to recognize each word, especially in poorer SNRs. To demonstrate the impact of word-level factors on assessments of listening difficulty, I will present the results of a study with 26 younger, normal-hearing adults. Participants recognized 160 monosyllabic words in noise from the Northwestern University Auditory Test Number Six (NU-6) in acoustically and cognitively demanding conditions. Multilevel models simultaneously assessed word-level (lexical frequency) and participant-level effects on a dual-task measure of listening effort. Results revealed interactions among acoustic, lexical, and cognitive demands on effort, even for items that were correctly identified. These findings highlight the importance of considering item-level difficulty on speech recognition tests even when using normed speech materials. [This work is supported by NIH/NIDCD R03DC015059.]