Electromyography artifacts are a well-known problem in electroencephalography studies [brain-computer interfaces (BCIs), brain mapping, and clinical areas]. Blind source separation (BSS) techniques are commonly used to handle artifacts. However, these may remove not only the EMG artifacts but also some useful electroencephalography (EEG) sources. To reduce this useful information loss, we propose a new technique for statistically selecting EEG channels that are contaminated with class-dependent EMG (henceforth called EMG-CCh). The EMG-CCh is selected based on the correlation between EEG and facial EMG channels. They were compared (using a Wilcoxon test) to determine whether the artifacts played a significant role in class separation. To ensure that the promising results are not due to the weak EMG removal, reliability tests were done In our data set, the comparison results between BSS artifact removal applied in two ways, to all channels and only to EMG-CCh showed that ICA, PCA, and BSS-CCA can yield significantly better ( ) class separation with the proposed method (79% of the cases for ICA, 53% for PCA, and 11% for BSS-CCA). With BCI competition data, we saw improvement in 60% of the cases for ICA and BSS-CCA. The simple method proposed in this paper showed improvement in class separation with both our data and the BCI competition data. There are no existing methods for removing EMG artifacts based on the correlation between the EEG and EMG channels. Also, the EMG-CCh selection can be used on its own or it can be combined with pre-existing artifact handling methods. For these reasons, we believe that this method can be useful for other EEG studies.