This randomised, double-blind, placebo-controlled study compared the efficacy of venlafaxine XR (75 or 150 mg/d) with diazepam (15 mg/d) over an 8-week treatment period in 540 non-depressed outpatients with generalised anxiety disorder (GAD). At week 8, significant improvements from baseline were observed in the venlafaxine XR, diazepam and placebo groups. Although these improvements were higher in the first two groups than in the placebo group for each of the primary efficacy variables (Hamilton Rating Scale for Anxiety (HAM-A) total, HAM-A psychic anxiety factor, Hospital Anxiety and Depression Scale (HAD) anxiety sub-scale and Clinical Global Impression (CGI) improvement), there were no statistically significant differences between groups. These non-positive results were thought to be due to the very high placebo response observed in some centres. To understand the variability of the study, a secondary preplanned analysis was performed. This involved sub-dividing the study centres according to their ability to detect a two-point mean difference between diazepam and placebo at week 8 on the HAM-A total score. Centres able to show such a difference were termed verum-sensitive. Improvements from baseline to week 8 in venlafaxine XR-treated patients from verum-sensitive centres were significantly greater than in placebo on each of the primary efficacy measures (P ≤ 0.05). This suggests that those centres able to detect an anxiolytic effect of diazepam were also able to detect an anxiolytic effect of venlafaxine XR. Significant differences in baseline demographics, rates of adverse event reporting and rates of patient discontinuations were noted between patients enrolled at verum-sensitive and verum-insensitive sites. These results reflect the importance of study centre selection in accurately determining efficacy in placebo-controlled trials.