Human reliability analysis plays a significant role in the industrial risk assessment as human errors often trigger the scenarios of emergencies or even major catastrophes. It is therefore very important to correctly estimate human error probabilities (HEPs). The paper considers the problem of obtaining a HEP within Human Cognitive Reliability (HCR) model which weights the abilities of individual professionals or teams to perform the task in a limited available time. Such a probability depends on performance shaping factors, in particular the factors characterizing individual’s knowledge and skills required for a certain task. Many circumstances stipulating effectiveness / non-effectiveness of human activity normally remain unknown or imperfectly known. This may not allow setting the distribution laws in the form of ordinary functions for random variables used in the HCR model. As an alternative, the authors apply the approach based on interval-valued (imprecise) probabilities technique oriented towards the sets of possible distributions. Accordingly, the resulting HEP is estimated as an interval of its possible values. In this framework, the lack of information is mitigated by involving a limited number of expert judgements. The new modification of the HCR model is more adequately representing real conditions since it requires fewer initial assumptions. This makes it possible to predict human reliability even under considerable uncertainty.