In a method for assessing the quality of an audio test signal derived from an audio reference signal by coding and decoding, the audio test signal is compared with the audio reference signal, as it were, behind the cochlea of the human ear. All masking effects as well as the transmission function of the ear are equally applied to the audio reference signal and the audio test signal. To this end, the audio test signal is broken down according to its spectral composition by means of a first bank of filters consisting of filters overlapping in frequency and defining spectral regions, said filters having differing filtering functions each determined on the basis of the excitation curve of the human ear with respect to the respective filter center frequency. The audio reference signal is also broken down according to its spectral composition into partial audio reference signals by means of a second bank of filters coinciding with the first bank of filters. Subsequently, a level difference by spectral regions is formed between the partial audio test signals and the partial audio reference signals belonging to the same spectral regions. To assess the quality of the audio test signal, a detection probability is determined, by spectral regions, on the basis of the respective level difference so as to detect a coding error of the audio test signal in the spectral region concerned.