The early weak fault features of gearbox is extracted difficult because of interference of disturbed signal, in view of this, a method of the full vector Stransformation is proposed to improve the differentiation degree of weak fault features. The double vertical channel vibration signals were integrated first with the full vector theory to guarantee the integrity of signal source information. Considering that the generalized S transform had the advantages of adaptive acquisition of the best time frequency spectrum according to timefrequency aggregation measurement criteria, then, the twodimensional (2D) timefrequency representation of fusion signals was realized through generalized S transformation. The fault features classifying working conditions were constructed with energy matrix in timefrequency series. The advantages of the full vector Stransformation in the extraction of weak fault features are verified by the experiments of the 20 groups of the wind turbine gearbox in the three kinds of weak faults, such as pitting, crack and uniform wear.