This article presents the results o f an evaluation and com parison study of three subjective techniques fo r determ ining human reliab ility under stress fo r w ork perform ed at tw o Taiwanese security companies. Stress levels at security com panies were estimated by using the Unified Tri-service Cognitive Performance Assessment Battery and 40 participants from tw o similar-sized companies. Experim ent results (m em ory searching task and m athem atical processing task) in d i cated that the participants were under high levels o f stress. Three subjective techniques (Success Likelihood Index M ethod, Technique fo r Human Error Rate Prediction, and Human Error Assess m ent and Reduction Technique) fo r estim ating human error probability were evaluated and com pared by using 20 experts fo r six tasks. The com parison criteria are interjudge consistency and accuracy. Of the three human error probability tests studied, the Technique fo r Human Error Rate Prediction and Success Likelihood Index M ethod were more consistent than the Human Error Assessment and Reduction Technique. The same relationship occurred in the com parison of accuracy. Thus, the Human Error Assessment and Reduction Technique m ay need to be m odified in some w ay if it is to be useful. It was already known that this technique required m odifications in error-producing conditions and nom inal human unreliability. Our w ork presents additional evidence to substantiate this. w ork stress human reliability analysis