The classical Split Hopkinson Pressure Bar technique is re-examined in order to optimize its accuracy, especially in the range of small strains. For many nonmetallic materials such as concrete, rocks, ceramics and polymers, the most important aspects of their behavior can be located in the range of small strains. The accuracy of the basic measurements of forces and velocities at both sample faces is discussed concerning the early stage of the loading. This accuracy depends on data processing which consists mostly of an accurate dispersion correction and of exact delays setting. A more precise wave dispersion correction and a new method to set exact origins of waves are then proposed. The validity of the average stress-strain curve obtained from measured forces and velocities is analysed using an one-dimensional numerical transient simulation of the tests. A fictitious specimen with a rate sensitive behavior described by a Sokolovsky-Malvern type constitutive model is used for this simulation. For the case where the classical SHPB analyses do not give acceptable results, an identification technique based on an inverse calculation method is presented. It relates material properties to forces and particle velocities measured at both faces of the specimen without using the assumption of axial uniformity of stresses and strains.