Electrode bridging is a common but easily ignored EEG artifact source. Based on the distinctive statistical characteristics of mutual information, a novel algorithm to automatically detect these bridges was developed and further applied to four EEG data sets acquired from different subjects. The applications identified four, four, three and zero pairs of bridged electrodes in these four data sets, respectively. No influencing factors were returned by One-way robustness analyses across different recording tasks and/or pre-processing procedures. And further comparison experiments performed on simulated data indicated that it outperformed the electrical distance method. All these findings suggest that the novel method is able to screen electrode bridges in a satisfying manner, making it of great significance in providing an indication to timely remedy the contaminated EEG data so as to avoid distortions to the resultant EEG topographies.