Background. In unselected samples, the diagnosis of major depression (MD) is not highly reliable. It is not known if occasion-specific influences on reliability index familial risk factors for MD, or how reliability is associated with risk for co-morbid anxiety disorders.

Methods. An unselected sample of 847 female twin pairs was interviewed twice, 5 years apart, about their lifetime history (LTH) of MD, generalized anxiety disorder (GAD) and panic disorder (PD). Familial influences on reliability were examined using structural equation models. Logistic regression was used to identify clinical features that predict reliable diagnosis. Co-morbidity was characterized using the continuation ratio test.

Results. The reliability of a LTH of MD over 5 years was fair (κ=0·43). There was no evidence for occasion-specific familial influences on reliability, and heritability of reliably diagnosed MD was estimated at 66%. Subjects with unreliably diagnosed MD reported fewer symptoms and, if diagnosed with MD only at the first interview, less impairment and help seeking, or, if diagnosed with MD only at the second interview, fewer episodes and a longer illness. A history of co-morbid GAD or PD is more prevalent among subjects with reliably diagnosed MD.

Conclusions. A diagnosis of MD based on a single psychiatric interview incorporates a substantial amount of measurement error but there is no evidence that transient influences on recall and diagnosis index familial risk for MD. Quantitative indices of risk for MD based on multiple interviews should reflect both the characteristics of MD and the temporal order of positive diagnoses.