Taking non-stationary and nonlinearity of epilepsy signals into consideration, we proposed a method for detection of epilepsy, based on Lempel-Ziv (LZ) complexity and empirical mode decomposition (EMD). EMD first decomposed epilepsy signals into a set of intrinsic mode functions (IMFs). Then calculated complexity of each IMF. Bonn dataset was utilized for evaluating the method. Experimental results showed that the highest accuracy could be achieved to 95. 25%. It has advantages of high accuracy, strong adaptability and so on.