Two experiments were conducted to evaluate the usefulness of limb tremor as an indicator of emotional arousal. In experiment 1, limb tremor measurements and subjective inventories were taken for two experimentally induced emotional arousal states (frustration and anxiety) and an experimentally induced relaxation state (calm) for two groups of 12 subjects each. A third group of 12 subjects had measurements taken for the relaxation state only. Tremor measurements were taken with an accelerometer, and the data were digitized and spectrally analyzed by computer. The subjective inventories were an abbreviated state form of the State-Trait Anxiety Inventory and a 7-point stress thermometer. The results of the experiment showed that, in a highly discrepant stimulus situation, the tremor and subjective measures were sensitive indicators of intra- and intersubject differences in emotional arousal. In addition, the stress thermometer data for two experimental groups correlated significantly with the limb tremor measures. In experiment 2, tremor measures and subjective inventories were taken for naturally occurring emotional arousal (anxiety) in 30 nonclinically anxious subjects before, during, and after an academic test-taking situation. Tremor measurements were taken with an accelerometer and the data digitized and spectrally analyzed by computer. The accelerometer was more sensitive than the one used in experiment 1. The subjective inventories were also more sensitive: the entire state form of the State-Trait Anxiety Inventory and a 30-point stress thermometer were used. The tremor and subjective measures failed to detect significant intersubject differences in emotional arousal across experimental sessions. The subjective measures also failed to detect any significant intrasubject differences, whereas the tremor measure did detect such differences for 14 out of the.'JO subjects. This demonstrates that tremor measure can be a sensitive indicator of intrasubject emotional arousal across experimental sessions. Another finding of experiment 2 proved significant. A Spearman rank correlation coefficient was computed to determine whether the spectrally analyzed variance estimates on the tremor data agreed with the variances computed on the raw data. The results (rs=.93, p<.001) suggest that the variance computed on the raw data itself would have served as a reasonable estimate of tremor changes across experimental sessions. An elaborate computer analysis of the data may be unnecessary, thus allowing a therapist to use a microcomputer program to compute quickly and inexpensively a variance on raw tremor data that would allow him to monitor a patient's changes in emotional arousal