Epilepsy is a chronic disease, and seizures are a limited episode of brain dysfunction caused by abnormal discharges of cerebral neurons. Seizures are characterized by periodic episodes and unpredictability. The automatic detection and analysis of seizures based on Electroencephalogram (EEG) data is of grate important for accurately identifying patients and reducing the work intensity of medical staff. This paper presents an epilepsy and seizure classification approach based on multi-spike liquid state machines. Firstly, this paper describes the network structure and neuron model, and introduces the EEG data encoding method. In addition, the proposed approach is used to solve the multi-classification problems of epilepsy data and conduct related experiments. At last, the experimental results show that the proposed approach has a great ability to solve the classification problems of epilepsy EEG data, and has a high classification accuracy rate. The average recognition accuracy rate reaches 71.23%, which has certain value in real life.