We show that the standard trust question routinely used in social capital research is importantly related to cooperation behavior and we provide a microfoundation for this relation. We run a large-scale public goods experiment over the internet in Denmark and find that the trust question is a proxy for cooperation preferences rather than beliefs about others' cooperation. To disentangle the preference and belief channels, we run a (standard) public goods game in which beliefs matter for cooperation choices and one (using the strategy method) in which they do not matter. We show that the "fairness question", a recently proposed alternative to the "trust question", is also related to cooperation behavior but operates through beliefs rather than preferences.