Bispectrum has been widely used to enhance the SNR. This is based on the assumption that the HOS properties of the signal of interest are different to the HOS properties of. the noise. In the present work, we consider the use of bispectrum techniques when repeated measurements are made of a deterministic signal embedded in random noise where SNR is in the range from -17 dB up to 0 dB. The performance of estimators and reconstruction algorithms are evaluated using simulated evoked potential data. We conclude that the best performance is achieved if we reconstruct the phase by bispectrum and amplitude through a second order method such as 'Spectral Subtraction'.