Abstract: The impact of model parameters of KICA on its optimization and separation performances is analyzed in theoretical aspects and demonstrated with experiments.In order to select proper model parameters,a time-efficient model selection method is put forward,which annexes a simple known signal to the sensor signals and gets optimal model parameter setting by minimizing the separation error.KICA with model selection step is applied to the task of removing ECG artifact from the EEG signal and the result shows KICA works effectively.