There is a need for a risk analysis technique specific for academic research laboratories. Since accurate accident data, normally required for quantitative risk analysis, are not available for this environment, expert judgements are often used to describe risks. However, these judgements are afflicted with linguistic, lexical or informal uncertainties. As a consequence, analyses made by different experts can lead to different results, which make risks incomparable. The purpose of this work is to analyse the effect of these uncertainties and to test strategies to improve the accuracy of the risk estimation based on expert judgements. Different calculation methods were used to compare the obtained risk scores. Results show that a multiplication-based formula, as used, for example, in the Failure Mode, Effects and Criticality Analysis (FMECA), has an inconsistent variance of the risk score distribution. Another approach, using a logarithm-sum-based formula, gives more consistent results but introduces other drawbacks. An estimation method based on Bayesian networks is giving more consistent variances, which are crucial for the risk estimation. With a higher precision of the risk score results, the prioritization of risks can be enhanced and resources can be better allocated to improve the level of occupational safety in academic research laboratories.