This paper represents an intersection between two lines of research. The first is portfolio choice theory, which underlies much of finance; the second is the elicitation of preferences under uncertainty. The theory of the behaviour of financial markets builds heavily on portfolio choice theory; until recently this has assumed that preferences are of a particularly simple kind. In contrast research on preferences has revealed that people have more sophisticated preferences. This paper tries to bring the two fields together by investigating, in a portfolio choice context, the preferences that are revealed by decisions. In the second of these two fields, researchers are increasingly using allocation problems to elicit the preferences of subjects, believing that such problems are more informative, and perhaps more natural, than other elicitation methods. At the same time portfolio choice theory is itself concerned with an allocation problem. Usually in experimental finance the allocation problems are over Arrow securities each of which pays off only in one state of the world. Instead we study the more realistic case, familiar from finance, in which all assets pay off in all states of the world. To make our study more realistic we frame the problem as one under ambiguity, where the probabilities of the states are not known to the decision-maker. This enables us to compare the performance of some recent theories of behaviour under ambiguity as well as traditional ones (such as Mean-Variance) from the theory of finance. We also identify a ‘rule of thumb’ that decision-makers may be using in this rather complex scenario. This research may help us to understand more fully actual portfolio choice decisions.