Each record of a bird on a survey is the result of three processes: (1) a bird being physically present, (2) this bird making itself available for detection, given that it is present, and (3) the bird being detected given that it has made itself available for detection. We employ a novel approach to examine the variation in availability for detection of the endangered Florida Grasshopper Sparrow (Ammodramus savannarum floridanus). Analyzing multi-day acoustic recordings, we quantified the effects of four sources of variation in the probability of singing: consistent variation among birds, day-to-day variation that was consistent from individual to individual, withinmorning variation consistent from individual to individual, and short-term variation related to birds' immediately previous state of singing (whether a bird had been singing in the immediate past). The two most important sources of variation were changes in singing probability with time of day and short-term differences due to an individual's intrinsic state, with birds currently singing being more likely to sing in the immediate future. The latter effect was not constant because nonsingers were more likely to start singing at morning civil twilight and singers were more likely to stop singing later in the morning. Monitoring Florida Grasshopper Sparrows needs to account for sharp drops in the probability of singing at times away from the daily window of peak singing activity. Additional analyses indicate that this conclusion holds across the species' range. We suggest that our findings have even wider applicability among birds.