This paper describes a technique that can predict electric fires by detection of discharge voltage signals caused by the use of electric facilities. In the experiment, various discharge modes, a flashover or a surface discharge through insulation paper and a line to line short, were simulated to acquire electrical information for predicting electrical fire as discharge modes. From the experimental results, it is hewn that electorial discharges which are ranked as majority causes of electric fires generate characterized signals distinguished from power frequency. Finally. We designed a prototype discharge detector based on the experimental results, and the detector is applied to a power lines. This study showed that the prediction of electric fires is possible by monitoring discharge voltage signals in electric power lines.