A reactive oxygen species (ROS) assay was previously developed for photosafety evaluation of pharmaceuticals, and the present multi-center study aimed to establish and validate a standard protocol for ROS assay. In three participating laboratories, two standards and 42 coded chemicals, including 23 phototoxins and 19 nonphototoxic drugs/chemicals, were assessed by the ROS assay according to the standardized protocol. Most phototoxins tended to generate singlet oxygen and/or superoxide under UV–vis exposure, but nonphototoxic chemicals were less photoreactive. In the ROS assay on quinine (200 µm), a typical phototoxic drug, the intra- and inter-day precisions (coefficient of variation; CV) were found to be 1.5–7.4% and 1.7–9.3%, respectively. The inter-laboratory CV for quinine averaged 15.4% for singlet oxygen and 17.0% for superoxide. The ROS assay on 42 coded chemicals (200 µm) provided no false negative predictions upon previously defined criteria as compared with the in vitro/in vivo phototoxicity, although several false positives appeared. Outcomes from the validation study were indicative of satisfactory transferability, intra- and inter-laboratory variability, and predictive capacity of the ROS assay. Copyright © 2012 John Wiley & Sons, Ltd.