A new method for numerical analysis of EEG is described. It consists of fitting the EEG with orthogonal polynomials in several steps to separate the different frequencies contained in the original recording (high-frequency noise, beta, sigma, alpha, theta, delta, low-frequency noise). The original EEG is separated into 5 different signals with cutoff frequencies at 40, 18, 4 and 0.6 Hz. The activity in these signals is classified into the usual EEG bands by zero-crossing. The results are given as integrated values over 1 s in the different EEG bands, or wave by wave, each wave being described by its time of occurrence, amplitude and duration. This method is designed for accurate amplitude measurement in the frequency bands of EEG, particularly for application in psychiatry and psychophys-iology.