Existing thermal comfort models defined in relevant standards are often found to be less effective for primary and secondary school students in educational buildings. This is often thought to be primarily due to differences in thermal sensation between children and adults. However, one important factor that is often neglected is the uncertainty associated with thermal comfort survey data. The existing method for calculating the mean thermal sensation vote is oversimplified and does not properly address related uncertainties. As a result, it ultimately affects the performance of the developed thermal comfort models. Hence, this research proposes a novel approach to compute the mean thermal sensation vote data for primary and secondary schools using Bayesian inference. This approach addresses the error caused by the uncertainties associated with the collected thermal sensation vote data in order to improve the effectiveness of the developed thermal comfort models for students. The proposed method was validated through a holistic case study using five thermal comfort models. The results showed that the accuracy of the developed thermal comfort models improved by 10.1% to 30.9%, and the R2 improved by 5.3% to 28.8%. A benchmark for the Bayesian model parameter setting was proposed as the reference for relevant studies. Finally, an open, user-friendly software was developed and is available to relevant users to implement the proposed approach more efficiently. The results of this research have practical implications for the development and optimization of thermal comfort models for students.