The Employment Retention and Advancement (ERA) Demonstration programme is a major current welfare-to-work social experiment, the largest random allocation evaluation ever mounted in Great Britain. This article draws on experience gained in designing the ERA Demonstration to explore the strengths and limitations of social experimentation for policy evaluation and analysis. The focus of the discussion is on the reasons for the choice of random allocation as a mean of estimating programme impacts, contrasting this approach with the alternatives. The weaknesses of random allocation designs are also examined in the light of the types of information policy-makers require from evaluations of labour market programmes and social policy demonstrations. The perennial ‘black box’ problem and the difficulties in generalizing from social experiments are given particular prominence.