The detection of multicomponent signals for which the components are not equidetectable is precisely investigated as a function of the level difference ΔLi/j between components. The detection thresholds are determined for a seven-tone complex signal with random starting phases masked by white noise. Level differences between the components are examined. A model for non-equidetectable conditions based on the statistical summation model is described. The improvement in detection is calculated from the level difference between components that is related to the thresholds for single components. The model predictions are in accordance with the experimental results.