Abstract. Recent studies related with earthquake prediction involve statistical studies of the ground electric self-potential behavior. Published results about the complexity of this kind of processes encourage us to study the statistical behavior of the ground electric self-potential recorded in Guerrero state, Mexico. This region is characterized by high seismicity. The electric self-potential variations were recorded in the Acapulco station directly from the ground. The sampling period was four seconds and the data were stored from March to December of 1993. Two significant earthquakes (EQs) occurred near this station, 15 May and 24 October whose magnitudes were Mw=6.0 and Mw=6.6 respectively. A preliminary processing was carried out consisting of a moving average of the original time series in order to filter the very high frequencies and to complete short lacks of data and outliers. Then, a visual inspection of the complete filtered signal was performed to search some seismic electric signals (SES), which were ambiguously depicted. Subsequently, a detrending of µ=0 was applied with the windows of 3.3, 6.6 and 10 h. Later, the analysis of the spectral exponent β was made, showing changes during the total period examined, and the most evident changes occurred during the preparation mechanism of the Mw=6.6 EQ. Fifteen days before the 24 October EQ, a Brownian-noise like behavior was displayed (β≈2), having a duration of about two days. In addition a Higuchi fractal method and wavelet analysis were made confirming the presence of the β-anomaly.