We experimentally study the effect of framing and size of large windfall gains on the redistribution of such gains. Randomly selected individuals from villages in Bangladesh were invited to take part in dictator experiments where they received endowments worth up to five months of average household income and were asked to distribute the endowment between themselves and other individuals. We manipulated whether dictators could GIVE to or TAKE from another individual (i.e. whether the endowment was allocated to the dictator or other individual) and whether the endowment was moderate (LOW) or very large (HIGH). We also provided dictators with the option to reconsider their original decision. We find that dictators allocate almost nine times more to other individuals under the TAKE than the GIVE frame when stakes are HIGH, even after they could reconsider their choices. In addition, we find that proportions allocated to other individuals dramatically drop when stakes increase under the GIVE but not the TAKE frame. The results provide novel evidence on the role of framing and stakes for pro-sociality.