Rock joints influence the stability of rock mass. Therefore, their shear strength is an important factor in determining the load a structure constructed on or in rock can withstand. Numerical and theoretical models are used to predict the shear mechanical behaviour of rock joints. These models are validated against data from laboratory testing. The data generation process from laboratory testing consists of several parts, each of which is associated with various sources of uncertainties. However, no model is better than the accuracy of the results it is validated against. Therefore, in this work, several approaches have been developed using data from a comprehensive experimental shear testing program, with the overall aim of reducing the uncertainties associated with various parts of the data generation process. Forty-six granite rock specimens containing both natural and artificially tensile induced joints were subjected to direct shear testing. Several tests were carried out for each parameter combination, allowing for statistical evaluations. The tests were conducted under controlled laboratory conditions under both constant normal stress and constant normal stiffness boundary conditions. These tests were performed under previously unexplored conditions, combining stresses occurring at depths of several hundreds of meters with a wide range of joint areas represented across three specimen sizes: 35 mm × 60 mm, 70 mm × 100 mm and 300 mm × 500 mm. In addition, fifteen replicas were manufactured from high-strength concrete based on one of the granite rock joints sized 70 mm × 100 mm for geometrical evaluations. Six of these replicas were subjected to direct shear testing and evaluated against the shear mechanical behaviour of the rock joint. Two quality assurance parameters for replica rock joints have been developed, which together with a method for establishing threshold limits of the parameters, reduces the uncertainties in parametric studies. An approach has been developed in which the effective normal stiffness is calculated and then inserted into the control system in tests under the constant normal stiffness boundary condition. The application of the effective normal stiffness essentially eliminates the error in applied normal load originating from the normal stiffness of the test system. Improved accuracy in displacement measurements has been achieved by applying optical displacement measurements directly over joint traces. Direct displacement measurements exclude errors in conventional measurements, which include undesired but unavoidable displacements originating from gaps and deformations in the test system. Approaches for consistent and physically based determination of both shear strength and shear stiffness have been developed. Analysis of variance shows that the shear strength of the tested rock joint specimens is not influenced by specimen size, whereas shear stiffness is. To sum up, the accuracy of data from laboratory tests is improved, constituting a prerequisite for improved models.