Spatio-temporal pattern analysis of EEG is an important tool in brain research. An EEG pattern analysis system based on a hierarchical multi-method approach is proposed here. The system consists of multiple steps including extraction of target signal, acquisition of intracranial electric activity distribution, adaptive segmentation of EEG and spatio-temporal pattern recognition. Some modern signal processing methods such as common spatial subspace decomposition, hidden Markov model are adopted. This paper also proposes an algorithm named LORETA-FOCUSS to estimate the current density inside the brain with a high spatial resolution. Microstate analysis of EEG is extended to the 3-D situation. The system was applied to the brain computer interface problem and achieved the highest accuracy of 88.89% with an average accuracy of 81.48% when classifying two imaginary movement tasks, while the data were not manually pre-selected. The result has proved spatio-temporal EEG pattern analysis is an efficient way in brain research.