Hydrofracturing is widely used for stress measurement because of its simplicity and high applicability to great depth. In the conventional hydrofracturing, supposing axi-parallel fracture plane in intact rock, the minimum stress σh in the plane normal to hole axis is calculated from the shut-in pressure Ps , and the maximum stress σH from breakdown pressure Pb or re-opening pressure Pr. But it has been pointed out that σH is less reliable because breakdown criterion depends on water permeation, and reopening is difficult to detect and vague in its mechanism. So it would be helpful if one could get information about σH from other than Pb or Pr. In this paper, the effect of stress difference (σH-σh ) on the probability distribution of the azimuth of vertical fractures is regarded as a new source of information about σH. Though the azimuth of fracture has been thought to coincide with that of maximum stress σH, one can easily expect that the fractures' azimuths would randomly scatter in direction when stress difference (σH-σh ) is small. The distribution of fracture azimuth, therefore, possibly provides the information about σH. In order to deal with such azimuth distribution caused by strength variation from location to location, this paper adopts the weakest link theory to investigate the breakdown phenomenon probabilistically. Based on the theory, the probability of breakdown and p. d. f of fracture azimuth are deduced, and the mean and 90 % confidence interval of Pb (or tensile strength T) and 90 % confidence interval of fracture azimuth are numerically calculated. From the results it was found that those calculated values are closely related to the stress difference or anisotropy providing new information about σH. This approach of stress estimation has been applied to in-situ test at a granitic rock mass with scarce joints.