This paper studies, both theoretically and experimentally, frame effects in the context of a public good game in which players have to make a costly contribution either (i) to achieve or (ii) not to lose a non excludable monetary prize. Our protocol leads to public good provision (not deterioration) only if a certain contribution level is achieved. Since both frames differ with respect to the reference point, we use Prospect Theory to derive testable predictions. In particular, Prospect Theory predicts more contribution in the second frame. Our evidence suggests that a) subjects' behavior is highly sensitive to frames and b) the theoretical prediction is confirmed except when the threshold is low. We also estimate the parameters which better suit our experimental evidence, partly confirming previous results in the literature.