Understanding human behavior from the perspective of normative and descriptive theories depends on human agents having stable and coherent decision-making preferences. Both utility theory (expected rational behavior; von Neumann & Morgenstern, 1947) and prospect theory with its certainty equivalent (CE) method (expected irrational behavior; Tversky & Kahneman, 1992) assume stable behavioral patterns of risk preferences. In contrast, our research pursues the opposite proposal: human preferences (rational or irrational) are not stable; variations in the decision context during risk elicitation determine people’s preferences even when the utilities of choice options are available. Accordingly, we found evidence that decision-makers reverse their risk preferences between CE tasks with logarithmically spaced certainty (unequal number of risk-averse and risk-seeking sure options) and linearly spaced certainty (equal number of risk-averse and risk-seeking sure options). The results revealed that the effect of probability range (low and high) on preferences, predicted by prospect theory, is an artifact of the logarithmically spaced sure options. When the sure options were linearly spaced the probability range no longer influenced risk preferences, indicating a preference reversal between decision tasks. Our findings highlight a need to investigate how the predictions of descriptive decision-making theories are shaped by their risk elicitation methods.