A Bayesian methodology is developed for groundwater leak detection at waste disposal sites. The method requires specification of the prior probability of a leak at the facility and updates this probability based on the proportion of constituent observations above a threshold value (e.g., the analytical detection limit) in the monitoring versus the background samples. The information content of a chemical measurement is characterized by the probability of occurrence of elevated concentrations of the chemical in a monitoring well, given the presence or absence of a facility leak. The evidence from multiple constituents is combined, assuming independence of errors, into a single aggregate estimate for the posterior probability of a leak. The procedure allows expert knowledge on site conditions and chemical behavior in the groundwater to be systematically incorporated in the regulatory evaluation of leak detection monitoring data. Applications using hypothetical and actual monitoring data from a hazardous waste facility demonstrate the procedure and highlight needs for further research.¿ 1997 American Geophysical Union