It has finally been nailed down:
This paper provides the first rigorous, empirical evidence of the existence of Giffen behavior, i.e., a situation in which consumers respond to an increase in the price of a good by demanding more of it. We begin by examining several theoretical approaches to the Giffen phenomenon and show that in each case Giffen behavior is closely associated with poor consumers’ need to maintain subsistence consumption in the face of an increase in the price of a staple commodity. We then present evidence on the existence of Giffen behavior among extremely poor households in two provinces of China. In order to obtain an unbiased estimate of the key price elasticity, we conducted a field experiment in which we randomly subsidized households’ primary dietary staple. Using consumption data gathered before, during and after the intervention, we find strong evidence of Giffen behavior with respect to rice in Hunan province.
The paper is here, hat tip Mark Thoma.
I find some of the authors' assertions a tad extreme: failure to empirically observe Giffen behaviour is not 'an indictment of neoclassical consumer theory', simply evidence that the theory extends to cover situations that are not frequently encountered in the real world.
All the same, this is a groundbreaking paper: even if Giffen behaviour is not neoclassical consumer theory's holy grail, these findings still constitute a small treasure.