Abstract: A method using variance as feature and using classification rate as one of evaluation criteria was proposed for the brain-computer interface (BCI) design of two kinds of imagery tasks. The wavelet theory was firstly discussed, and cross-banding of wavelet packet decomposition was analyzed. Variances of wavelet and wavelet packet coefficients were taken as features, then the two EEG features were extracted from the electrodes C3 and C4, and they were finally classified by using a linear support vector machine. The results showed that the maximum classification accuracies of both features were 86.43% and the corresponding times were 4.32 and 4.31 s. So, it was suitable to use wavelet variance and wavelet packet variance as features. The presented classification rate could reflect the classification accuracies and classification time at the same time, and also give a new idea for classification of imagery tasks in BCI.