A method is provided for discarding an audio message if the audio message meets a predefined discarding condition, which may factor in at least the audio message's total length, presence of spoken word(s) in the audio message, presence of tone(s) in the audio message and the presence of click(s) in the audio message. An audio message may be discarded if no spoken words are detected in it and the audio message's total length is shorter than a predefined minimal message length, or it contains a tone and the audio message's total length minus the tone's length is shorter than the predefined minimal message length, or it contains a click sound and the audio message's total length minus the click's length is shorter than the predefined minimal message length. An audio message may be first stored in a memory buffer and then moved to a legitimate messages pool memory or to a spurious messages pool memory. A messages discriminator and a voicemail system that utilize the messages discrimination method are also provided.