Value Surveys may reveal well-behaved societies by the statistical treatment of the agents’ declarations of compliance with social values. Similarly, the results of experiments conducted on games with conflict of interest trace back to two important primitives of social capital – trust and reciprocity – which can be used to explain deviations from the Nash equilibrium and which lead to the optimal cooperative outcome. In this paper we attempt to elicit the true motive(s) underlying the behaviour of players in experimental trust and dictator games and suggest that the most informative utilization of surveys in this regard goes beyond the simple comparison of answers to a questionnaire with actual behaviour. Specifically the paper uses descriptive statistics and ordered probit models to analyse whether, and to what extent, answers to a questionnaire about attitudes to trusting and reciprocating predict subjects’ behaviour and, by comparing behaviour in Trust and Dictator Game, disentangles the strategic and altruistic motivations. We find no simple or direct correlation between behavioural trust or trustworthiness and attitudinal trust or disposition to reciprocate. However, dividing subjects according to attitudinal trust and trustworthiness, we observe that the link between the questionnaire and experimental sessions is more subtle than the mere correlation between average attitudes and average behaviours. The information conveyed by a survey appears to be much more powerful ex post – once the two motivational components have been separated out.