A multisite, double-blind, placebo-controlled study was made of 177 patients with the diagnosis of generalized anxiety disorder. After a 1-week placebo lead-in, they were randomized to 4 weeks of treatment by placebo or one of two doses of a novel, nonsedating compound that had demonstrated reduction of fear-avoidance behavior in animals. Efficacy was not demonstrated at a significant level. A further analysis of 142 patients who completed the treatment was undertaken to test the hypothesis that efficacy might be demonstrated by the single-rater procedure (SRP), which eliminates interrater error variance. The 80 patients who were examined by the same clinical rater for all six visits were compared with the 62 patients who had the multiple-rater procedure (MRP), ratings by two or more clinicians sequentially over the six visits. A two-way analysis of variance showed significantly greater discrimination of placebo and drug for the MRP group. The results provided no support for the frequent preference for the SRP. The MRP may include less psychotherapeutic interaction with the patient and less researcher bias in ratings.