Acoustical correlates of stress can only be evaluated in comparison with some "standard" specifying which syllables are actually stressed. The Standard should be consistent from time to time, and largely independent of talker and listener idiosyncrasies. Three phonetically-trained subjects listened to repeatedly spoken texts and spontaneous sentences until they could categorize each syllable as either stressed, unstressed, or reduced. This procedure was repeated three times for each speech text and listener. Two listeners differed from each other on only 5% of all syllables as to whether they were perceived as stressed or not. Each showed about 5% confusion in decisions about stressed syllables from one trial to another. Unstressed and reduced levels were confused more frequently. The third listener gave less consistent results. Subjects' judgements of stress when given only the written text were of comparable consistency but did not correspond well with perceptions with speech, if the speech was spontaneous rather than spoken texts. Stress perceptions consequently may be suitable for evaluating acoustical correlates to within a 5% tolerance in overall location scores. Pooling the perceptions from several trials and several listeners may improve the stability of this "standard" for stress assignment. (Author/DD)