Experimental studies of biotic interactions in real field conditions are essential to un-derstand the structure and functioning of ecological networks. The use of artificial caterpillars to mimic actual prey availability is generally seen as a standard approach to compare the activity and diversity of predators along environmental gradients. Yet, even with standardized material and procedures, biases may still affect data from multiple observers with different expertise. We used pictures of artificial caterpillars with or without various predation marks, in an online survey that was targeted for the participants of the project, to evaluate the reliability of predation marks identifica-tion made by non- scientists and by scientists with and without experience in preda-tion mark identification. Pictures of artificial caterpillars displayed typical marks left by birds, mammals, and arthropods, as well as non-predation marks (‘false positive’). In total 357 respondents scanned 7140 pictures of artificial caterpillars. Scientists were more confident and accurate in their observations than non- scientists, but the differences in correct identifications among scientists and non- scientists were low. Scientists with experience were also more accurate than scientists without experi-ence, whereas there were no differences in self-confidence among scientists with and without experience. Accuracy in predation mark identification did not differ among types of predators, but respondents were more keen to identify marks left by birds or mammals than marks left by arthropods. Our results have practical implications for the design of multi-observer projects relying on artificial caterpillars as a proxy to as-sess predation intensity, in particular in the context of citizen science.< Réduire