Empirical evidence suggests that disturbance has profound effects on the species diversity of aquatic and terrestrial assemblages. Conceptual ecological theories, such as the intermediate disturbance hypothesis (IDH), predict maximum diversity at intermediate levels of disturbance. Tests of the predictive power and generality of these models are, however, hampered by the fact that the meaning and units of "disturbance" are not clearly defined. For example, it is seldom recognized that the rate of disturbance is the product of both frequency and extent (e.g., area or volume) of disturbance events. This has important consequences for the design and interpretation of experiments on disturbance. Here we present, for the first time, an experimental design that allows for unconfounded testing of combinations of area and frequency (i.e., regimes) for a given rate of disturbance. We tested the prediction that species richness responds differently to equal rates of disturbance, depending on the specific combination of frequency and area, on marine hard-substratum assemblages. Five different rates of disturbance and two regimes (small frequent or large infrequent disturbances) were applied at three sites. The results showed that the effect of a certain rate of disturbance (1) varies strongly among assemblages and (2) also depends on the specific combination of frequency and area of disturbance events. Maximum species richness was observed at intermediate rates of disturbance at site 1 (i.e., support for the IDH), whereas there was a monotonic decline at site 2 and there was no evident pattern at site 3. The variable responses among sites were explained by differences in degree of competitive exclusion and rates of recruitment. At the site where the IDH was supported, the regime with a large proportion of the area disturbed infrequently showed higher richness, compared to the regime with a small proportion disturbed frequently. This was likely due to a stronger decrease of dominants, which allowed for the recruitment of new colonizing species. In summary, we conclude that tests and general syntheses of models of disturbance-diversity patterns would benefit from more explicit definitions of the components of disturbance, as well as a stronger focus on the importance of variation in inherent properties of natural assemblages.