The invention discloses a multi-scale binary tree blast furnace fault diagnosis method based on sample segmentation, belonging to the technical field of blast furnace fault diagnosis. The method comprises the following steps of: firstly, acquiring blast furnace production condition and equipment operation state data, detecting the data and performing normalization for the extracted data through a mean-variance normalization method; converting a blast furnace fault diagnosis problem into a dichotomy problem to perform multi-classifier design; finding a segmentation plane through an improved generalized eigenvalue support vector machine, converting into two dichotomy problems, respectively finding a distance measuring matrix having local properties and being adaptive for each type of fault data itself, and designing two classification hyperplanes based on different scales through the support vector machine. The method provided by the invention is suitable for identification of high-dimensional nonlinear fault data; and by means of segmenting sample data and measuring similarity among the samples with a multi-scale standard, the method gives consideration to global and local logic structures of the identified data, reduces complexity of the identified fault problem and improves precision of fault diagnosis.