Frustration is used as a criterion for identifying usability problems (UPs) and for rating their severity in a few of the existing severity scales, but it is not operationalized. No research has systematically examined how frustration varies with the severity of UPs. We aimed to address these issues with a hybrid approach, using Self-Assessment Manikin, comments elicited with Cued-Recall Debrief, galvanic skin responses (GSR) and gaze data. Two empirical studies involving a search task with a website known to have UPs were conducted to substantiate findings and improve on the methodological framework, which could facilitate usability evaluation practice. Results showed no correlation between GSR peaks and severity ratings, but GSR peaks were correlated with frustration scores -- a metric we developed. The Peak-End rule was partially verified. The problematic evaluator effect was the limitation as it confounded the severity ratings of UPs. Future work is aimed to control this effect and to develop a multifaceted severity scale.