In a variety of biological and physical phenomena, temporal fluctuations are found, which are not explainable as consequences of statistically independent random events. If these fluctuations are characterized by a power spectrum density S(f) decaying as f -¢ at low frequencies, this behaviour is called 1/f noise.Counting statistics applied to earthquake activity data leads to three time scales with different characteristics, represented by the exponent/3: at interval lengths less than 1 h, the shocks are randomly distributed as in a Poisson process. For medium time intervals (1 day to 3 months), the exponent 1 +/3 is larger (1.4 for M0=3), but approaches unity for higher threshold magnitudes M0. In longer time ranges the exponent assumes values near 1.55, however, with increasing statistical variation at higher M0, due to lower counts.The temporal sequence is different from white noise; thus, it might be fruitful to apply neural network algorithms, because this method allows predictions in some other cases with similar characteristics.