We propose a new method for detecting the onset of the stretch reflex response for assessment of spasticity based on the Tonic Stretch Reflex Threshold (TSRT). Our strategy relies on a three-stage approach to detect the onset of the reflex EMG activity: (i) Reduction of baseline activity by means of Empirical Mode Decomposition; (ii) Extraction of the complex envelope of the EMG signal by means of Hilbert Transform (HT) and; iii) A double threshold decision rule. Simulated and real EMG data were used to evaluate and compare our method (TSRT-EHD) against three other popular methods described in the literature to assess TSRT ('Kim', 'Ferreira' and 'Blanchette'). Four different groups of signals containing simulated evoked stretch reflex EMG activities were generated: groups A and B without spontaneous EMG activity at rest and signal-to-noise ratio (SNR) of 10dB and 20dB respectively; groups C and D with spontaneous EMG activity at rest, as observed frequently in spastic muscles, and SNR of 10dB and 20dB respectively. The results with simulated data showed a significantly higher accuracy of TSRT-EHD for detecting the onset of the reflex EMG activity in groups C and D when compared to the other methods. Analyses using real data from five post stroke spastic subjects demonstrated that the TSRTs generated by each method were dramatically different from one another. Nevertheless, only TSRT-EHD provided valid measures across all subjects.