Five food-deprived rhesus monkeys were exposed to 225-MHz continuous-wave, and 1.3-GHz, and 5.8-GHz pulsed radiation to determine the minimal power densities affecting performance. The monkeys were trained to press a lever (observing-response) thereby producing signals that indicated availability of food. In the presence of the aperiodically appearing food signals, a detection response on a different lever was reinforced by a food pellet. Continuous, stable responding during 60-min sessions developed and was followed by repeated exposures to radiofrequency radiation. The subjects, restrained in a Styrofoam chair, were exposed to free-field radiation while performing the task. Colonic temperature was simultaneously obtained. Observing-response performance was impaired at increasingly higher power densities as frequency increased from the near-resonance 225 MHz to the above-resonance 5.8 GHz. The threshold power density of disrupted response rate at 225 MHz was 8.1 mW/cm2; at 1.3 GHz it was 57 mW/cm2, and at 5.8 GHz it was 140 mW/cm2. These power densities were associated with reliable increases in colonic temperatures above sham-exposure levels. The mean increase was typically in the range of 1 degree C, and response-rate changes were not observed in the absence of concomitant temperature increases. In these experiments increase of colonic temperature was a much better predictor of behavioral disruption than was either the power density of the incident field or estimates of whole-body-averaged rates of energy absorption.