The anticipation of major events and, more generally, the making of optimally accurate predictions represent permanent challenges for decision-makers in both the political and economic fields. Various techniques and models exist for achieving this, having been developed over a great many years. And yet the occurrence in recent times of events that virtually no one had predicted –such as the 9/11 attacks– has lent fresh impetus to research in this area. The book by Philip Tetlock and Dan Gardner presented in this article by Charles du Granrut is part of this movement of new research on forecasting. Based on a programme called “The Good Judgment Project”, and on the results of a forecasting tournament organized by an agency within the US intelligence services (the main lessons of which it outlines), it enables us to identify the profiles and methods of the best forecasters –the so-called superforecasters– and proposes a number of rules for refining the quality of forecasts to the greatest possible degree. Charles du Granrut summarizes the essential points here, while stressing some of the limitations relating, among other things, to the field being studied (geopolitics), the short time-horizon and the relatively simple wording of the propositions considered (which call only for binary answers).