Prestige-biased social learning occurs when individuals preferentially learn from others who are highly respected, admired, copied, or attended to in their group. This form of social learning is argued to reflect novel forms of social hierarchy in human societies, and, by providing an efficient short-cut to acquiring adaptive information, underpin the cumulative cultural evolution that has contributed to our species’ ecological success. Despite these potentially important consequences, little empirical work to date has tested the basic predictions of prestige-biased social learning. Here we provide evidence supporting the key predictions that prestige-biased social learning is used when it constitutes an indirect cue of success, and when success-biased social learning is unavailable. We ran an online experiment (n=269) in which participants could copy each other in real-time to score points on a general-knowledge quiz. Our implementation of prestige was the number of times someone had previously been copied by others. Importantly, prestige was an emergent property of participants’ behaviour during the experiment; no deception or manipulation of prestige was employed at any time. We found that, as predicted, participants used prestige-biased social learning when the prestige cue was an indirect cue of success, and when direct success information was unavailable. This highlights how people flexibly and adaptively employ social learning strategies based on the reliability of the information that such strategies provide.