Background: Mediastinal staging in patients with non-small cell lung cancer (NSCLC) with endobronchial ultrasound-guided fine-needle aspiration (EBUS-FNA) requires a high negative predictive value (NPV) (ie, low false negative rate). We provide a conservative calculation of NPV that calls for caution in the interpretation of EBUS results. Methods: We retrospectively analyzed our prospectively gathered database (January 2007 to November 2011) to include NSCLC patients who underwent EBUS-FNA for mediastinal staging. We excluded patients with metastatic NSCLC and other malignancies. We assessed FNAs with rapid on-site evaluation (ROSE). The calculation of NPV is NPV = true negatives/true negatives + false negatives. However, this definition ignores nondiagnostic samples. Nondiagnostic samples should be added to the NPV denominator because decisions based on nondiagnostic samples could be flawed. We conservatively calculated NPV for EBUS-FNA as NPV = true negatives/true negatives + false negatives + nondiagnostic. We defined false negatives as negative FNAs but NSCLC-positive surgical biopsy of the same site. Nondiagnostic FNAs were nonrepresentative of lymphoid tissue. We compared diagnostic performance with the inclusion and exclusion of nondiagnostic procedures. Results: We studied 120 patients with NSCLC who underwent EBUS-FNA; 5 patients had false negative findings and 10 additional patients had nondiagnostic results. The NPV with and without inclusion of nondiagnostic samples was 65.9% and 85.3%, respectively. Conclusions: The inclusion of nondiagnostic specimens into the conservative, worst-case-scenario calculation of NPV for EBUS-FNA in NSCLC lowers the NPV from 85.3% to 65.9%. The true NPV is likely higher than 65.9% as few nondiagnostic specimens are false negatives. Caution is imperative for the safe application of EBUS-FNA in NSCLC staging.