Results are presented of a collaborative study between 19 European laboratories on the variability of the Ames test. Examples are shown of various methods that are generally used to evaluate an Ames test without reference to a specific statistical model: the number of revertants per plate, mutation factors (increase over the spontaneous value) and determination of the doubling concentration. Considerable variations between test results occurred, between laboratories as well as within laboratories. Partly this was due to different interpretations of the guidelines given, as these allowed some flexibility. The results were also influenced by other factors, some of which are perhaps not yet generally recognized. Apart from the level and quality of the S9 preparations, the most important factor might be the number and physiological condition of the cells plated. When the results from all experiments were considered together, 60-80% of the test results were found within the half- to two-fold range of the overall median. This might be considered satisfactory for a study not using rigidly standardized test conditions. From the experience with the present study, several recommendations are given for the design and performance of future collaborative studies.