Social norm interventions represent a low-cost and effective policy tool that have been shown to generate behaviour change in a number of contexts. We investigate whether valence framing impacts the effectiveness of a social norm intervention on prosocial behaviour. We use Amazon Mechanical Turk in conjunction with oTree to conduct an experiment in which we manipulate descriptive beliefs and original endowments in the context of a dictator game. We find that the impact of a social norm intervention appears to be significantly greater in a frame of negative valence vs. a frame of positive valence. Loss aversion and positional preferences could play a role in these findings. Regression results furthermore indicate a differences in the determinants of allocation decisions across frames, suggesting that normative beliefs matter more in a positive frame, whereas descriptive beliefs matter more in a negative frame. This work contributes to a better understanding of framing effects as well as the conditions under which normative considerations are most salient. On a practical level, it points to an opportunity for policymakers to substantially improve upon the design of social norm interventions