The discretization method starts from the elementary single value intervals and then searches for the best merge between adjacent intervals. Two different types of merges are encountered. First, merges with at least one interval that does not meet the constraint and second, merges with both intervals fulfilling the constraint. The best merge candidate (with the highest chi-square value) is chosen in priority among the first type of merges (in which case the merge is accepted unconditionally), and otherwise, if all minimum frequency constraints are respected, among the second type of merges (in which case the merge is accepted under the condition of improvement of the confidence level). The algorithm is reiterated until both all minimum frequency constraints are respected and no further merge can decrease the confidence level. When compared with other chi- square based methods like ChiMerge and ChiSplit methods, this global evaluation carries some intrinsic benefits. The Khiops automatic stopping rule brings both ease of use and high quality discretizations. Its computational complexity is the same as for the fastest other discretization methods.