Purpose <p>Exhaled gas analysis is a non-invasive and straightforward method for the diagnosis of lung cancer. This study aimed to assess whether characteristic compounds in exhaled gas can serve as specific diagnostic markers for lung cancer.</p> Methods <p>This study enrolled 410 participants, comprising 102 healthy individuals and 308 lung cancer patients. Participants' exhaled gases were collected using the in-house fabricated exhaled gas collection device. The exhaled gases were then detected using extractive electrospray ionization mass spectrometry (EESI-MS). Subsequently, partial least squares discriminant analysis (PLS-DA) was conducted to screen for significantly different compounds. A lung cancer diagnostic model was constructed through the support vector machine (SVM) algorithm.</p> Results <p>A total of 17 and 11 characteristic compounds common to cohort 1 and cohort 2 were obtained by PLS-DA analysis in both positive and negative ion modes. A lung cancer diagnostic model was constructed based on Cohort 3. The model contained 17 characteristic compounds in positive ion mode with 92.50% accuracy and 92.39% sensitivity. In negative ion mode, the model with 11 characteristic compounds also performed excellently with an accuracy of 90.83% and a sensitivity of 94.57%. Cross-validation in Cohorts 1 and 2 confirmed the model's robustness, with area under the curve values surpassing 0.97 and both accuracy and sensitivity rates exceeding 90.00%. Ultimately, 28 significantly differential metabolites were identified.</p> Conclusions <p>This study underscores the potential of exhaled gas analysis as a reliable method for diagnosis of lung cancer.</p> Trial registration <p>ClinicalTrials.gov NCT06086587, registered on 26 September 2023.</p>