Observing and quantifying fish behaviour towards active sampling gears like trawls is challenging. A high-frequency imaging sonar (DIDSON) was applied to directly observe and record fish avoidance behaviour in front of the mouth of an active pelagic surface trawl in the meso- to eutrophic Czech reservoir Želivka in August 2009. In this multi-species environment the fish displayed a variety of behavioural patterns. We classified 11 different categories of avoidance (n=1051) based on fish reaction and swimming behaviour towards the active trawl using Sonar5-Pro software. Classification of these categories was based on eight continuous variables: (1) total fish length, (2) distance to the trawl, (3) tortuosity, (4) mean speed of total track, (5) mean speed of first avoidance track, (6) mean speed of total avoidance track, (7) angle between initial and first avoidance track and (8) angle between initial and total avoidance track. The acoustic analysis showed that 76% of the fish in trawling area was captured and 24% swam out of it. A higher proportion of the fish showed an avoidance reaction during the day (44%) than during the night (6%), indicating that ambient light conditions trigger this behaviour during the day. A two-step cluster analysis of the tracks revealed three distinct clusters that corresponded to the most abundant fish species in the reservoir: ‘bleak-cluster’, ‘bream-cluster’ and ‘silver carp-cluster’.