This paper addresses the traditional Lakonishok, Shleifer and Vishny (LSV) herding measure and points out its lack of internal consistency. Frey, Herbst and Walter (2007) have shown by empirical simulations that LSV is biased. We provide a formal explanation of this bias and propose a more appropriate measure of herding. We then turn to the properties of the new indicator proposed by Frey, Herbst and Walter (2007): we show that it is accurate only under very strong assumptions, and propose a corrected version of their indicator. We also show that the real herding value is within an interval bounded by LSV and FHW. Finally, as both corrected measures require a prior knowledge of some parameters of the distribution, we conclude that measuring accurately herding intensity is a much more difficult task than considered up to now in the empirical literature.