This paper describes a resource for the study of spontaneous speech under stress, a corpus of 216 unscripted task-oriented dialogues conducted by normal Canadian adults in the course of a sleep deprivation experiment under 3 drug conditions. Speakers carried out the route-communication task (see [1]) in alternation with a battery of other tasks over a 6-day study which included a 60-hour sleepless period. Each speaker participated in 12 dialogues. The design permits comparisons within speakers for sleep deprivation (baseline, deprived, post-recovery), and between speakers for drug condition (placebo, d-amphetamine, Modafinil) and number of conversational partners encountered (1, 2). Preliminary examination of dialogue length, task performance, and aspects of dialogue strategy indicate effects of all these variables. Effects of sleep-deprivation and drug condition are less severe than those found in simpler tasks [7].