Abstract: In this study, an algorithm of P300 waveform extraction and target classification was proposed based on temporal-delayed and spatio-temporal filtering. Firstly, the multi-channel electroencephalogram (EEG) signal was delayed in temporal domain. And a cost function was constructed based on the least square method. The alternately optimizing was conducted to estimate the spatio-temporal filter and the desired signal until the cost function was converged. At last, the spatio-temporal filter could be obtained to separate the components in the spatial domain and extract the P300 waveform in the temporal domain. And then, simulation analysis was carried out to verify the waveform extraction performance of the algorithm with P300 data. The results show that the algorithm is better than the correlative algorithm for P300 waveform recovery. Finally, the obtained spatio-temporal filter was utilized to extract P300 components as classification features from real EEG data. A Fisher linear discriminant analysis was trained with the P300 components got from training dataset and utilized to classify the EEG signals. The results indicated that the P300 waveform extraction performance, classification accuracy rate and area under curve (AUC) value of the proposed algorithm are significantly better than the correlative algorithm. Therefore, the proposed algorithm can extract P300 waveform and classify target effectively.