The purpose of the experiment was to study how the loudness (or annoyance) of the constituents in pairs of community noises combines into total loudness (or total annoyance). Three community noises (pile driving, jack hammering, street traffic) were combined pairwise at different sound levels. The observers judged the total loudness of the combined noises as well as the loudness of each component noise when heard alone. Three models of loudness (or annoyance) summation for noise were tested: a vector summation model, a model assuming that the loudnesses of the masked constituent noises add arithmetically, and a simple model stating that the total loudness equals the loudest of the component noises when heard alone. All three models fitted the data satisfactorily from a statistical point of view. The ’’loudest component’’ model was favored because it produced a prediction error of only 14% in the ’’worst case’’ and can, therefore, serve as a ’’rule of thumb’’ for many practical purposes.