<p>Exhaled volatile organic compounds (VOCs) have been investigated in some diseases, including cognitive impairment, in pilot studies. The present study aimed to explore the role of exhaled VOCs in identifying and differentiating patients with Alzheimer’s disease (AD). The identification cohort included 241 participants: 99 AD patients (dementia=74, mild cognitive impairment (MCI) = 25), 59 non-AD dementia patients, and 83 cognitively unimpaired controls (CUCs). Proton transfer reaction time-of-flight mass spectrometry (PTR-TOFMS) was employed to detect exhaled VOCs. The differences in VOCs between the AD and CUC groups and between the AD dementia and non-AD dementia groups were compared separately. Furthermore, machine learning models for discriminating AD from CUC as well as AD dementia from non-AD dementia were established. The AD diagnostic model was further validated in an independent cohort of 44 AD patients (dementia=33, MCI = 11) and 35 CUCs. Moreover, we explored the possible metabolic pathways of AD-specific exhaled VOCs with the Human Metabolome Database (HMDB). We detected 60 different VOCs between the AD and CUC groups, among which the top ten were C₄H₁₀S, C₂H₆N₂O₂, C₃H₆O₃, C₄H₇F₂NO, C₆H₅NO₂, C₈H₆N₂, C₃HN, C₉H₂₁N, C₁₅H₂₄, and C₈H₁₂ (<i>P</i> &lt; 0.001). The AD diagnostic model had an accuracy of 0.93 in the identification cohort according to internal validation and 0.75 in the validation cohort for external validation. The model for discriminating AD dementia patients from non-AD dementia patients had an accuracy of 0.90. Fourteen of the 60 AD-specific VOCs were retrieved from the HMDB, from which the three most significant metabolic pathways were identified, namely, butyrate metabolism, pyruvate metabolism, and glycolysis/gluconeogenesis. These results indicate that exhaled VOC measurement may be a promising approach for diagnosing and differentiating AD, but further validation is needed.</p>