Abstract: A decision maker elicits expert predictions to estimate the probability of an uncertain event. Experts are recruited sequentially to report costly advice. Optimal sample size is achieved when the marginal value of advice equals to the marginal cost. However, marginal value of a prediction is difficult to judge without knowledge of the information structure. We propose a stopping rule for expert sample. Each new expert reports a second-order prediction on the average prediction of previous experts, as well as their prediction on the event. Recruitment stops when the marginal value of second-order predictions drops below the marginal cost. We show that our stopping rule is a good proxy for optimal stopping in a variety of information structures and reputational concerns of experts.