Hydroacoustic data acquired for estimating fish populations contain information on both fish and non-fish targets, so sonar technicians traditionally rely on their knowledge of fish behaviour and experience with hydroacoustics to remove non-fish targets from the hydroacoustic data. This process is often labour-intensive and time-consuming, making real-time assessment of fish populations difficult. Simple solutions are not always available for all circumstances. However, the split-beam sonar data collected in the lower Fraser River, British Columbia, Canada, showed distinct signatures between actively swimming fish and non-fish objects such as drifting debris, surface bubbles, and stationary objects in the water column and off the river bottom. Acoustic tracks of fish and non-fish targets were characterized by differentiable statistical patterns that were amenable to discriminant function analysis (DFA). An application of DFA to segregate fish and non-fish targets detected by a split-beam sonar system in the lower Fraser River is presented, characteristics of user-identified fish and non-fish acoustic tracks being utilized as learning samples for the DFA. Also, a method to rank the discriminating power of individual variables is presented, providing guidance for constructing efficient and effective discriminant functions with variables that offer high discriminating power. The DFA yielded classification accuracies of 96% for fish and 91% for non-fish tracks and reduced the manual sorting time by 50–75%.