A problem of screening agents with privately known forecasting abilities and reservation utilities and then eliciting truthful information from these agents once they are employed is considered. For the case of risk averse agents and large principals, an approximately optimal solution is constructively characterized under the assumption that payoffs are normally distributed. A significant extension of the deFinnetti-Savage probability elicitation result is developed. Under this extension knowledge of agent preferences is not required (when the underlying conditional distributions are symmetric) and this fact is exploited to solve the screening problem in a mean-variance formulation.