Most formal decision theories are deterministic. But people’s actual behaviour shows variability: when a participant in an experiment is presented with exactly the same decision tasks more than once within a short space of time, he/she is liable to answer at least some of them differently from one presentation to another. Why does this happen? Can seemingly systematic departures from standard (expected utility) theory be partly/largely explained accordingly? What is the status of axioms – e.g. independence, transitivity – and notions such as ‘attitude to risk’ and ‘attitude to ambiguity’ when preferences are probabilistic? Can we make sense of ‘framing effects’? What are the implications for policy? Too many questions to be answered in a single seminar, but I will introduce some basic framework and talk a little about a couple of recent experiments.