The extraction of highly discriminant features is crucial for successful species identification of fish shoals if backscattered narrowband signals do indeed contain discriminant information. Four different methods of feature extraction are described and applied to the same data, providing new descriptors expected to improve species identification. Echograms, amplitude probability density function (PDFs) and spectral features are used to describe acoustic images of single shoals. Image processing is used to improve signal shoal description, by taking into account the shoal structure and species-related spatial distribution. Three pelagic species are considered: sardine (Sardina pilchardus (Walbaum)), anchovy (Engraulis encrasicolus (L.)), and horse mackerel (Trachurus trachurus (L.)) detected during fisheries acoustics surveys conducted in the Bay of Biscay. A correct classification rate of 57% overall was found for data covering a mesoscale oceanographic environment and including seasonal variability. If space and time scales are reduced this value increased to 98%, emphasizing the value of non-acoustic and a priori information.