This article reports the results of a meta-study of 96 prisoner's dilemma studies comprising more than 3,500 participants. I disentangle the role of 'risk' (to co-operate unilaterally) and 'temptation' (to defect against a co-operator) and find that: (i) an index of risk best explains the variation in cooperation rates across one-shot games, while (ii) an index of temptation best explains the variation in finitely repeated games. Risk and temptation indices also affect gender comparisons. Women are more co-operative than the average man if risk is low and less co-operative if risk is high. There are no gender differences on average. 2 Andreoni and Croson (2008) summarise some of the research on the 'partner' versus 'stranger' question in a handbook article on the public good game. In the prisoner's dilemma Andreoni and Miller (1993) and Cooper et al. (1996) found more co-operation in the 'partner' condition, Boone et al. (1999) found no difference and Andreoni (1988) found more co-operation in the stranger setting. Boone et al. (1999) find more co-operation in the 'partner' condition with a long horizon in the repeated game and no difference with a short horizon of the repeated game. See also the more detailed discussion of differences between these papers in Section 4. Most articles on the prisoner's dilemma I found do not contain data on both the 'partner' and the 'stranger' case. The findings on gender are consistent with evidence in Simpson (2003), who argues precisely that the reason why existing literature on the prisoner's dilemma has, by and large, not found gender differences is because of the presence of both temptation and risk in these games (Kuwabara, 2006). 3 To the extent that communication reduces strategic uncertainty, this could explain for example why preplay communication is often found to increase co-operation rates in these settings (see e.g. the meta study of Balliet, 2010or Ledyard, 1995and Chaudhury, 2011 for surveys on the related public good game). Other examples of interventions designed to reduce temptation could include taxation schemes, with very progressive schemes better suited to curb temptation compared to linear schemes.