We present a social learning experiment in which subjects predict the value of a good in sequence. We elicit each subject's belief twice: first ("first belief"), after he observes his predecessors' prediction; second, after he also observes a private signal. Our main result is that subjects...