With a growing demand for hydroelectric energy, the number of reservoirs is dramatically increasing worldwide. These new water bodies also present an opportunity for the development of fishing activities. However, these reservoirs are commonly impounded on uncut forests, resulting in many immersed trees. These trees hinder fish assessments by disrupting both gill-netting and acoustic sampling. Immersed trees can easily be confused with fish schools on echograms. To overcome this issue, we developed a method to discriminate fish schools from immersed trees. A random forest algorithm was used to classify echo-traces at 120 and 200 kHz, recorded by an EK80 (SIMRAD) in narrowband (continuous wave) and in broadband mode (frequency modulated). We obtained a good discrimination rate between trees and schools, especially in broadband (90% ratio of good classification). We demonstrate that it is possible to discriminate fish schools from immersed trees and thus facilitate the use of fisheries acoustics in reservoirs.