In central Europe forests, the share of European beech (Fagus sylvatica L.) trees has been increased in the last decades. Machine strength grading of hardwood is challenging due to a lack of knowledge about strength predictors. However, high strength classes are needed for the utilization as glued and cross laminated timber. We used the information of an industrial scanner on fiber orientations, developed a 3D cluster value (SOG_3D,150,max) for strength assessment and combined the parameter with the dynamic modulus of elasticity (MOE_dyn) to compute an indicating property (IP). A sample of 407 European beech boards passed a multi-sensor scanner to detect wood density, eigenfrequency and slope of grain (SOG). Fiber angle on the surfaces of four board sides was measured using the tracheid effect. The spatial fiber orientation inside the board was modeled for a total of approximately 150,000 points per board meaning 12 points per cm³. Finally, the board section with the largest average local fiber orientation in a window of 150 mm defined the grading parameter SOG_3D,150,max. The prediction of tensile strength via SOG_3D,150,max reached r² between 0.466 and 0.605 depending on the type of data transformation. A combination with the MOE_dyn, the probably most common IP, increased the r² to 0.722 at best. Local grain deviation is a suitable wood parameter for hardwood strength grading. By detecting local defects, the causality between wood strength and tree functioning as well as silvicultural steering may be further understood in future.