The study of the brain bioelectrical activities in terms of deterministic nonlinear dynamics is quite recent [1]-[3]. Perspectives of clinical use of the nonlinear tools range from an augmented predictive classi cation of Alzheimer's desease [4] to the possibility of controlling epileptic activity [5]. However, the nonstationarity of the brain states plus the poor Signal-to-Noise Ratio (SNR) due to the large amount of brain processes active at the same time have forbidden so far a generally accepted procedure to estimate suitable indicators [2], [6], [7]. Whence the need for a suitably tailored lter able to extract the activity under study to avoid contamination between noise and bioelectrical activities while maintaining a su cient number of data to perform indicator calculations. The reason for this communication is twofold. We apply for the rst time a novel ltering technique based on the local features of the wavelet transformation [8] and, second, we propose a procedure for brain signal processing, making use of available nonlinear tests. The procedure here suggested estimates the presence of determinism in a magnetoencephalogram using the newly developed test [9], then, whenever allowed, it calculates the correlation dimensions (CD) [10] and tests the validity of the calculation. We will show that, while the measured signal contains deterministic activity, it is not possible to reasonably estimate the CD, for which we nd, however, an upper bound of 7.