Major depressive disorder (MDD) is increasingly to be recognized as a chronic, deteriorating illness with the high risk to obtain comorbidity. In order to provide clinicians with a subjective approach to decide appropriate treatments for MDD patients, a real time detection system for predicting the antidepressant responses is important. Wavelet Transform and five nonlinear methods Largest Lyapunov Exponent (LLE), Detrended Fluctuation Analysis (DFA), Fractal Dimension (FD), Correlation Dimension (CD) and Approximate Entropy (ApEn) were applied to extract the features from electroencephalography (EEG) activities in antidepressant responses. Non-parametric analysis, correlation analysis and confusion matrix were employed to evaluate the performance of classifying and decide the optimal threshold for discrimination. Moreover, the system is built to aid clinicians' in prediction of the antidepressant responses before treatments and the results can be viewed within 40 seconds.