We estimate the causal effect of carbon footprint labels on individual food choices and quantify potential carbon emission reductions, using data from a large-scale field experiment at five university cafeterias with over 80,000 individual meal choices. Results show that carbon footprint labels led to a decrease in the probability of selecting a high-carbon footprint meal by approximately 2.7 percentage points with consumers substituting to mid-carbon impact meals. We find no change in the market share of low-carbon meals, on average. The reduction in high-carbon footprint meals is driven by decreases in sales of meat meals while sales of mid-ranged vegan, vegetarian and fish meals all increased. We estimate that the introduction of carbon footprint labels was associated with a 4.3% reduction in average carbon emissions per meal. We contrast our findings with those from nudge-style interventions and discuss the cost-effectiveness of carbon footprint labels. Our results suggest that carbon footprint labels present a viable and low-cost policy tool to address information failure and harness climatarian preferences to encourage more sustainable food choices.