Depth of anaesthesia is hard to define and not readily measurable. Recently, attention has tumid to evoked potentials (EPs) rather than the electroencephalogram (EEG) and they have been validated as a good measure of depth of anaesthesia. However, the amplitudes of the EPs vary from tenths of a microvolt to a few microvolts (mu V). and are embedded in the spontaneous EEG waveform whose amplitude is typically 10 to 30 mu V. Thus. in most instances the signal-to-noise ratio (S/N) is less than 1:10 (-20 dB). It is this small signal-to-noise ratio that makes waveform and signal estimation classification difficult. Therefore. an intelligent signal processing methodology for evoked potentials in anaesthesia monitoring and control is proposed in the paper. A model-based algorithm based upon auto-regressive with exogenous input (ARX) models is used to improve the signal-to-noise ration (S/N). Quantitative feature extraction is implemented to extract the factors describing the changes in amplitudes and latencies of the mid-latency auditory evoked response. In this way, three principal factors are obtained and then merged together using qualitative fuzzy logic to create a reliable index for monitoring depth of anaesthesia. Twenty-one clinical trials were carried out to validate this methodology as a reliable assessment of anaesthetic depth during intravenous anaesthesia using propofol.