Lecturer promotion and tenure decisions are critical both for university management and for the affected lecturers. Therefore, they should be made cautiously and based on reliable information. Student evaluations of teaching quality are among the most used and analysed sources of such information. However, to date little attention has been paid in how to process them in order to be able to estimate their reliability. Within this paper we present an approach that provides estimates of such reliability in terms of confidence intervals. This approach, based on Bayesian inference, also provides a means for improving reliability even for lecturers having a low number of student evaluations. Such improvement is achieved by using past information in every year's evaluations. Results of applying the proposed procedure to university-wide data corresponding to two consecutive years are discussed.