Time-domain (TD) features have been widely used in pattern recognition-based control systems. The TD features, zero crossing (ZC), slope sign change (SC), and Willison amplitude (WAMP) make use of thresholds to attenuate background noise. Inconsistent thresholds have been reported in the literature and relatively little work has been done to investigate the effect and robustness of thresholding. Therefore, the aim of this study was to develop a novel method for investigation of thresholding TD features based on classification performance. Experiments including multi-channel surface electromyography recordings during hand movements were carried out for three separate days. The effect and robustness of thresholding TD features were assessed by scatter matrix separability criterion (SMSC), support vector machine (SVM) classification, and statistical tests. Results obtained for identical thresholds for ZC, SC, and WAMP ranged between 0.67 μV and 1.76 μV for all channels and days. An interval recommended for future threshold investigation of a factor r, ranging between 0 and 0.52, was identified. Furthermore, results revealed that thresholds were not robust over a period of six days. This indicates that investigation of thresholding TD features should be performed for each specific application. The recommendation is to use the method, introduced in this study, for investigation of thresholding TD features in future applications.