We evaluated the effectiveness of the cocaine test for diagnosing Horner's syndrome. The test was administered to 119 patients with a diagnosis of Horner's syndrome and to 50 normal subjects. We compared the cocaine-induced anisocoria in the two groups by measuring photographs of the pupils. We found the cocaine test to be highly effective in separating normal subjects from patients with Horner's syndrome. The chances of having Horner's syndrome increased with the amount of cocaine-induced anisocoria. Through the use of logistic regression analysis, we determined the odds ratio of having Horner's syndrome compared with not having it for each 0.1-mm increment of anisocoria measured after cocaine administration. A postcocaine anisocoria value of 0.8 mm gave a mean odds ratio of approximately 1050:1 that Horner's syndrome was present (lower 95% confidence limit = 37:1). We found that simply measuring the postcocaine anisocoria provided a better prediction of Horner's syndrome than taking the trouble to calculate the net change in anisocoria. Odds ratios should help the clinician decide if the result of a cocaine test is indicative of Horner's syndrome.