Abstract: This paper discusses the classification of seabed sediment using normally incident, high frequency sound pules. A number of investigators used reflectivity of sound on the sea bottom and some other quantities as features for classification of sediment types. We suggest a new feature which is independent upon reflectivity and roughness of seabed. Our feature is extracted from relative form of echo envelope, it strongly correlates to attenuation of sound in sediments. The attenuation of sound is nearly proportional to frequency and at our frequency (120 kHz) the attenuation in different sediments varies in the range of about tenfold. Therefore we think the attenuation is a good feature for classification. In order to avoid the influence of surface backscattering, we used a directive transducer put near to the bottom. In laboratory a Bayes classifier was carried out by computer, and a fairly high success rate has achieved.