BACKGROUND: Olfactory abnormalities are reported in Alzheimer's disease and Parkinson's disease. Anosmia appears to be common in dementia with Lewy bodies but not in pure Alzheimer's disease. OBJECTIVE: To determine whether anosmia improves discrimination between the Lewy body variant (LBV) of Alzheimer's disease and "pure" Alzheimer's disease. METHODS: 106 cases of necropsy confirmed pure Alzheimer's disease (n = 89) or LBV (n = 17) were reviewed. All had received butanol odour threshold testing. Anosmia was defined as a score < or = 1.0 on a 0-9 point scale. Logistic regression analysis was used to model potential predictors (for example, parkinsonism, smoking, hallucinations) of neuropathological diagnosis and anosmia. RESULTS: LBV cases had an increased prevalence of anosmia (65%) compared with Alzheimer's disease (23%; odds ratio (OR) = 6.3, p = 0.00045), or normal elderly people (6.7%). Within the dementia cases, the negative predictive value (92%) and specificity (78%) of anosmia were both good; sensitivity for detecting LBV was 65%, but the positive predictive value (PPV) was only 35%. Logistic regression models showed anosmia (OR = 5.4, p = 0.005) and visual hallucinations (OR = 7.3, p = 0.007) were strong independent predictors of Lewy body pathology. When anosmia was added as a core feature to consensus diagnostic criteria for probable Lewy body dementia, five additional cases of LBV were detected (29% increased sensitivity), but with four additional false positives (1% increased discrimination, 4% decreased specificity, 33% decreased PPV). CONCLUSIONS: Anosmia is very common in LBV. Adding anosmia as a core feature improved sensitivity for detecting LBV, but did not improve discrimination between Alzheimer's disease and LBV owing to a concomitant increase in false positives.