We presented a new method for stress measurements which is based on the hypothesis that fractures which are critically-stressed for frictional sliding in the current stress field tend to permeable and those which are not critically-stressed tend to be impermeable. Fracture depth, dip and strike can be detected from borehole imaging logs such as BHTV/UBI or FMI/FMS. By comparing the depths of fractures and those of thermal anomalies detected on temperature logs, we can identify which of the fractures are permeable. Stresses on fracture planes can be estimated from the fracture orientation provided that the state of in-situ stress is assumed. Therefore, knowledge of which fracture orientations tend to be permeable and hence support critical stress states, and which fracture orientations are not allows us to make a grid search over possible states of stress to identify the stress state that best satisfies the observations and hypothesis. A simple description of the stress states is necessary. Assuming that one principal stress is vertical and equal to the overburden and pore pressure is known, the stress states are defined by three unknowns given by the two depth gradients of the two horizontal stresses, and the orientation of the maximum horizontal stress. The range of admissible stress states can be further reduced by considering the strength of the bulk rock. We applied the method to estimate stress state about a 3.6 km deep well in granite at the Soultz Hot Dry Rock test site. Fractures and thermal anomalies detected in the depth range of 2850∼3500 m were used for the estimation. Results indicate that the stress regime is one of normal/strike-slip faulting, consistent with the fault plane solutions of induced and regional earthquakes. The predicted profile of minimum principal stress magnitude agrees with that determined from the hydrofracture method. The inferred orientation of the maximum principal stress differs from the best available estimate (defined by thermally-induced tension fractures) by 20∼30°. The example demonstrates that the method can yield useful estimates of the state of stress at great depth from relatively common data of temperature, borehole imaging, and density logs without requiring any additional input from expensive test. The method can also be applied in high temperature environments where conventional approaches to stress estimation cannot be applied.