In this project we propose a measure of social capital based on the behaviour in a public good game. We play the public good game within 28 groups in two similar neighborhoods in Cartagena, Colombia, one of which had been targeted for over two years by a conditional cash transfer program that has an important social component. The level of cooperation we observe in the 'treatment' community is considerably higher than in the 'control' community. The two neighborhoods, however, although similar in many dimensions, turned out to be significantly different in other observable variables. The result we obtain in terms of cooperation, however, is robust to controls for these observable differences. In the last part of the paper we also compare our measure of social capital with other more traditional measures that have been used in the literature.