To evaluate the efficacy of endoscopic brush cytology in diagnosing transitional cell carcinoma of the upper urinary tract. Sixty-three endoscopic brush cytology specimens from 48 patients were compared with corresponding cytologic specimens obtained by irrigation and catheterization as well as histologic specimens. Twenty patients (25 brushes) had histologically documented transitional cell carcinoma (TCC) or carcinoma in situ (CIS) of either the ureter or renal pelvis. Among these, 8 (32%) of the brush samples were reported as positive for TCC, 10 (40%) atypical or suspicious, and 7 (28%) negative. The seven negative cases were ultimately shown to be low grade (I-II/IV) TCC. Combining atypical and positive diagnoses, the calculated sensitivity for diagnosis of TCC by this technique was 72%. The irrigations or catheterized urines from these same patients yielded lower sensitivity, 48%, and detected only higher grade lesions. Ten patients were proven histologically to have nonneoplastic disease (hydroureter, obstruction, inflammation). Sixteen of the 17 brush specimens from these patients were negative, resulting in a specificity of 94%. In the remaining 18 patients (21 brushes) there were 17 negatives and 4 atypicals. Concomitant cytology supported the brush diagnosis in all but one sample. Brush cytology is a specific and more sensitive sampling method than irrigation or catheterized urine in detecting TCC of the upper urinary tract. Brush cytology does not appear to be successful in diagnosing dysplasia or CIS. As with urinary cytology in general, the technique is less effective in diagnosing low grade (I and II) lesions.