Are larger groups better at cooperation than smaller groups? This paper investigates, under controlled conditions, the presence and direction of a possible group size effect in pure public good provision by large heterogeneous groups. Employing subjects drawn from the general population and introducing Internet-based procedures to study this question, we collected experimental evidence from 1,110 subjects playing a linear public goods game in groups of 10, 40, and 100 members. We find a positive and significant group size effect: Increasing group size by a factor of 10 (4) increased efficiency by 10 (6) percent. The effect arose at the intensive margin and with repetition. Those who contributed contributed more in larger groups. Larger and smaller groups had similar initial contribution levels, but cooperation rates declined more slowly in the larger groups. Free-riding was invariant to group size, despite subjects' persistent beliefs of a negative group size effect at the extensive margin. Further econometric examination of the data supports these findings and provides starting points for future theoretical an d experimental research on the group size effect.