Hydraulic fracturing is extensively used in hydrocarbon extraction and in Engineered Geothermal Systems. However, many aspects involved in the initiation, propagation and interaction of existing and newly-created fractures are not clearly understood. Since the data obtained from real-scale stimulations provide solely indirect information on the fracturing processes, laboratory and numerical analyses play a crucial role for better understanding the mechanisms involved in the hydraulic fracturing of rocks. This paper compares the observations made in hydraulic fracturing experiments performed on prismatic granite specimens into which two flaws were precut, with finite element analyses. Both the experiments and numerical analyses used specimens with the double-flaw geometry 2a-30-30 to which 1) only water pressure inside the flaws or 2) water pressure inside the flaws and an external vertical load were applied. Two stages of crack development, namely crack initiation and crack coalescence, and the maximum and minimum principal stresses around one of the inner flaw tips, were assessed. When no vertical load was applied, the hydraulic macro-fractures observed experimentally are tensile, and initiate from the tip end of the studied flaw and propagate to coalesce in the bridge between inner flaw tips. The numerical model results are consistent with these observations, since high tensile principal stresses occur at the tip end of the flaw and in the bridge between flaws. On the other hand, when a vertical load of 5 MPa was applied, the hydraulic macro-fractures observed experimentally are tensile but initiate from the upper face of the flaw tip and propagate in a vertical direction and not towards the bridge between inner flaw tips. This is also consistent with the numerical results, since maximum tensile stresses only occur at the upper face of the flaw, and both maximum and minimum principal stresses are compressive in the bridge between inner flaw tips, a condition which would not lead to the development of tensile cracks.