One of the strongest determinants of behavioural variation is the tradeoff between resource gain and safety. Although classical theory predicts optimal foraging under risk, empirical studies report large unexplained variation in behaviour. Intrinsic individual differences in risk-taking behaviour might contribute to this variation. By repeatedly exposing individuals of a small mesopredator to different experimental landscapes of risks and resources, we tested 1) whether individuals adjust their foraging behaviour according to predictions of the general tradeoff between energy gain and predation avoidance and 2) whether individuals differ consistently and predictably from each other in how they solve this tradeoff. Wild-caught individuals (n = 42) of the jumping spider Marpissa muscosa, were subjected to repeated release and open-field tests to quantify among-individual variation in boldness and activity. Subsequently, individuals were tested in four foraging tests that differed in risk level (white/dark background colour) and risk variation (constant risk/variable risk simulated by bird dummy overflights) and contained inaccessible but visually perceivable food patches. When exposed to a white background, individuals reduced some aspects of movement and foraging intensity, suggesting that the degree of camouflage serves as a proxy of perceived risk in these predators. Short pulses of acute predation risk, simulated by bird overflights, had only small effects on aspects of foraging behaviour. Notably, a significant part of variation in foraging was due to among-individual differences across risk landscapes that are linked to consistent individual variation in activity, forming a behavioural syndrome. Our results demonstrate the importance of among-individual differences in behaviour of animals that forage under different levels of perceived risk. Since these differences likely affect food-web dynamics and have fitness consequences, future studies should explore the mechanisms that maintain the observed variation in natural populations.

Most values were collected from recorded videos using the program EthoVision (Noldus).