This work reports the results of two online experiments with a general-population sample examining the performance of different tasks for the elicitation of risk attitudes. First, I compare the investment task of Gneezy and Potters (1997), the standard choice-list method of Holt and Laury (2002), and the multi-alternative procedure of Eckel and Grossman (2002) and evaluate their performance in terms of the number of correctly-predicted binary decisions in a set of out-of-sample lottery choices. There are limited differences between the tasks in this sense, and performance is modest. Second, I included three additional budget-choice tasks (selection of a lottery from a linear budget set) where optimal decisions should have been corner solutions, and find that a large majority of participants provided interior solutions instead, casting doubts on people’s understanding of tasks of this type. Finally, I investigate whether these two results depend on cognitive ability, numerical literacy, and education. While optimal choices in budget-choice tasks are related to numerical literacy and cognitive ability, the predictive performance of the risk-elicitation tasks is unaffected.