This paper presents an experimental study of the random incentive mechanisms which are a standard procedure in economic and psychological experiments. Random incentive mechanisms have several advantages but are incentive-compatible only if responses to the single tasks are independent. This is true if either the independence axiom of expected utility theory or the isolation hypothesis of prospect theory holds. We present a simple test of isolation in the context of choice under risk. In the baseline (one task) treatment, we observe risk behavior in a given decision problem. We show that by adding an asymmetrically dominated choice problem in a random incentive mechanism risk behavior can be manipulated systematically; this violates the isolation hypothesis. The random incentive mechanism thus does not elicit true preferences in our example.