A two-phase survey of mental disorders uses a screening test to identify possible cases, thereby reducing the resources devoted to interviewing those not having the condition of interest. It is demonstrated using a mathematical model that in situations likely to be encountered in practice a two-phase design may lead to an increase in the efficiency of prevalence rate estimation, and also to an improvement in the efficiency of case detection. However, in certain applications the modest gain in efficiency may not warrant the additional complexity of a two-phase approach to data collection. Data from a survey of mental disorders in Edmonton, Canada, which collected information on 3258 community residents using the Diagnostic Interview Schedule and the General Health Questionnaire, are used to demonstrate how two-phase methods would have changed the efficiency of an actual survey.