Disturbance cues are released by stressed or disturbed prey prior to a predator attack and convey useful risk assessment information regarding local threats. While studies have shown that disturbance cues may be important early on within the predation sequence (prior to an attack), their role in predator–prey interactions remains relatively overlooked by ecologists. Critically, experimental studies examining disturbance cues, especially among prey fishes, have been conducted primarily under laboratory or semi‐natural conditions. Here, we tested the prediction that disturbance cues function as sources of risk assessment information in situ. We exposed Trinidadian guppies, in two natural populations differing in predation risk, to a model predator paired with stream water or the disturbance cue collected from guppies from either a high‐ or low‐predation risk population. We found that the predator inspection response of guppies to disturbance cues depends on the level of risk of both the focal and the cue source population. Guppies from both populations exhibited increased latencies to inspect, lower inspection rates and reduced inspecting group sizes towards the model paired with conspecific disturbance cues versus a stream water control. Interestingly, guppies of both populations showed evidence of higher perceived predation risk towards the disturbance cues collected from high‐predation risk donors compared to low‐predation risk donors. Our results support the hypothesis that disturbance cues function as a source of information used by prey fish in the assessment of predation risk and provide the first evidence of disturbance cue function under fully natural conditions.