a abdi - ntelligent Systems Group (ISG). Dept. of Plec & Comp. Eng. Tehran University P. O. Bor ۱۴۳۹۵/۵۱۵, Tehran, Iran
m allamehzadeh - Seismology Division, International Institute of Earthquake Enginecring and Seismology (IIEES)P.O Box ۱۹۳۹۵/۳۹/۳, Tehran, Iran
c lucas - ntelligent Systems Group (ISG). Dept. of Plec & Comp. Eng. Tehran University P. O. Bor ۱۴۳۹۵/۵۱۵, Tehran, Iran Seismology Division, International Institute of Earthquake Enginecring and Seismology (IIEES)P.O Box ۱۹۳۹۵/۳۹/۳, Tehran, Iran Intelligent System
m bahavar - Seismology Division, International Institute of Earthquake Enginecring and Seismology (IIEES)P.O Box ۱۹۳۹۵/۳۹/۳, Tehran, Iran

Neural relworks are powerful icecrigent and computational tools, proposed and developed in the hope of achieving human-like performance. They have been very successful ir. solving many practical engineering problems. Ir the statistical pattern recognilion context, they are optimum nonlinear classifiers in the Bayesian sense. They can form complex and non-planar decisioa surfaces belween populations with sigpificant overlaps. In tliis paper, we have investigated ihe problem of tclcscismic source discrimination based on the short period records, using the probabilistic neural network. In addition to the known and useful discriminant features like the P and P-coda spectra, we have also supplied our neural classifier with the noise information. Our major conclusion is that incorporating the effect of noise through the S/N spectriin, along with the use of neural network classifiers, cam yield better discrimination resulis than the conventional features and classilicrs.