This is the first in a series of reports on the long-term test-retest reliability and procedural validity of the UM-CIDI, a modified version of the Composite International Diagnostic Interview used in the US National Comorbidity Survey (NCS). This report focuses on DSM-III-R Generalized Anxiety Disorder (GAD). The NCS administered the UM-CIDI to a nationally representative sample of 8098 respondents in the age range 15-54. A subsample of 36 respondents was subsequently selected for clinical reappraisal of GAD, consisting of reinterviewing by a clinical reappraisal interviewer who blindly readministered the GAD section of the UM-CIDI followed by an expanded version of the GAD section of the Structured Clinical Interview for DSM-III-R (SCID). The test-retest reliability of UM-CIDI/DSM-III-R lifetime GAD is Kappa = .53. When the requirement that the worries be excessive or unrealistic (A2) is removed, as in ICD-10 and partially in DSM-IV, reliability increases to Kappa = .78. The concordance between the baseline UM-CIDI diagnosis and the SCID diagnosis is Kappa = .35, while the cross-sectional concordance is Kappa = .47 (.66 when the Criterion A2 requirement is removed). Item-level analysis shows that lack of concordance between the UM-CIDI and the SCID is due largely to Criteria A2 and D. The A2 problem could be addressed either by deemphasizing the cognitive-evaluative component of GAD as in ICD-10, or by removing consideration of the term "unrealistic" from the criterion as in DSM-IV and more clearly specifying the meaning of the term "excessive". These options require further research on similarities and differences in risk factors, course, family history, and treatment response of more narrowly and broadly defined GAD. The Criterion D problem is due to lack of clarity in what constitutes a symptom occurring "often". This is clarified in DSM-IV. It is likely that this clarification will make it possible to develop more precisely structured questions to evaluate Criterion D in subsequent revisions of the UM-CIDI, resulting in improved reliability and validity.