Fiberoptic bronchoscopy (FB) has a low yield in the diagnosis of chronic cough (greater than 3 weeks) in unselected patients. We assessed the yield of FB for cough during a four-year period in patients with nonlocalizing chest roentgenograms who were refractory to diagnostic efforts and empiric bronchodilator or antitussive therapy. Seven (28 percent) of 25 patients undergoing FB for cough (of greater than 1,500 bronchoscopies) had diagnostic findings (broncholithiasis, two; tracheobronchopathia osteochondroplastica, two; and tuberculous bronchostenosis, laryngeal dyskinesia, and arytenoid polyp, one each). No tracheobronchial neoplasms were detected. Age greater than 50 years and female sex independently predicted positive results (p = 0.02 Fisher's exact test), while duration of cough (two to 240 months), airflow, and smoking status did not. When patients with prior pulmonary or extrathoracic neoplasms were excluded, seven (35 percent) of 20 studies were diagnostic. Diagnoses potentially could have been made by thoracic computed tomographic scanning in four patients and indirect laryngoscopy in two. Fiberoptic bronchoscopy has a respectable yield for diagnosis of refractory chronic cough and is a reasonable procedure in carefully selected patients.