The purpose of the present study was to examine the test-retest reliability of the alcohol and drug modules of the AUDADIS-ADR in three sites: Bangalore, India, Jebel, Romania and Sydney, Australia. The overall reliability of ICD-10, DSM-IV and DSM-III-R dependence diagnoses was found to be good to excellent for each substance, including alcohol, for each time frame, regardless of whether the total sample or user subsample figured into the calculations. Reliability associated with corresponding harmful use and abuse diagnoses were mixed, but generally lower. Reliability statistics for Bangalore were generally lower than those of the Jebel and Sydney sites, particularly for alcohol diagnostic criteria. Implications of these results are discussed, in conjunction with results from the discrepancy interview protocol analyses within sites, in terms of future revisions to the AUDADIS-ADR and its training procedures tailored to developing countries.