Randomised control trials (RCTs) have attracted increasing attention in evaluation of public policy. They have become some sort of gold standard to determine “what works” in international development. However, in some cases, the key feature of RCTs – the random allocation of treatment(s) – may be unethical, politically infeasible, or simply impossible. This paper discusses the main barriers to answer causal questions of key interest in policy evaluation. It then reviews approaches complement to RCTs which can be used to answer evaluation questions in conjunction with RCTs and alternatives useful whenever RCTs are not applicable. The paper discusses issues in drawing causal inference from these alternative approaches and provides some examples from specific sectors where standard RCTs are less likely to be applicable.