Traditional Support Vector Machine (SVM) is widely used classification method for brain-computer interface (BCI). However, SVM has a high computational complexity. In this paper, Gaussian Mixture Model (GMM)-based training data reduction is proposed to reduce high computational complexity. The proposed method is configured as follows: First, wavelet-based combined feature vectors are applied for motor imagery electroencephalography (EEG) identification and principal component analysis (PCA) are used to reduce the dimension of feature vectors. Thereafter, the GMM is implemented to reduce training data sizes. Finally, a nonlinear SVM classifier is used to classify the reduced training data. The performance of the proposed method was evaluated using three different motor imagery datasets in terms of accuracy and training time. The results from the study indicate that the proposed method achieves high accuracy with faster computational time in motor imagery EEG classification.