In the original specification of cumulative prospect theory, distinct sets of parameters control for the curvature of the value function and the shape of the probability weighting function. There is one for the gain domain and one for the loss domain. However, in most estimations, behaviour over losses is assumed to perfectly reflect behaviour over gains, through a unique set of parameters. We examine the consequences of relaxing this simplifying assumption in the context of Tanaka et al.’s (2010) risk-experiment procedure. On the one hand, we show that subjects’ behaviour for gains is mostly reflected for losses at the aggregate and individual levels, and is consistent with the cumulative prospect theory fourfold pattern. However reflection is partial as the mean curvature of the value function is slightly less convex for losses than it is concave for gains. These results are robust to a high-stake context. Then, we demonstrate that assuming reflection when measuring loss aversion is innocuous neither at the aggregate nor at the individual level. On the other hand, we highlight the existence of a strong, negative and persistent framing effect on values elicited for loss aversion.