Abstract We compare sensitivity and reliability of three basic techniques of shear-wave splitting analysis – cross-correlation of wave components, minimization of covariance matrix eigenvalues and minimizing energy on the transverse component – of both synthetic and real data, point out some problems and propose solutions. We discuss major difficulties of evaluating shear-wave splitting parameters, which are not always obvious in routine processing and can sometimes lead to false results and wrong interpretations. We describe effects of pre-processing of digital records of shear waves, e.g., filtering and setting lengths of analyzed signals, and discuss reliability of results to minimize outliers in evaluated splitting parameters. For reliable evaluation of splitting of medium-quality SKS wave forms, which are valuable mainly in temporary deployments of stations, we recommend a careful band-pass filtering of the signal, using a window length not shorter than the dominant period, inspecting the signal/noise ratio on T component and the particle motion. The transverse minimization method provides usually more robust and stable solutions than the other two methods. In the conclusion, we suggest how to proceed in evaluation of splitting parameters to ensure reliable solutions even for medium- or low-quality shear waveforms recorded frequently during temporary field measurements.