Abstract: In order to study critical slowing down characteristics of surface electric potential (EP) before unstable failure of roof rocks, loading failure experiment of rock specimens was carried out, the surface EP signal of rocks was collected in loading process. Based on theory of critical slowing down, variance and autocorrelation coefficient of the potential signal time sequence were calculated and analyzed. The analysis results are as follows: there is a good correspondence between the change of surface potential, the load and strain during the whole loading process; EP signal of rock has the phenomenon of critical slowing down in the loading process, the variance and autocorrelation coefficient of EP has sudden increase and continuously increasing trend before the main rupture, it can be used as a precursory signal to indicate unstable failure of rocks; the different window lengths and lagging steps affect the volatility stability and variation tendency of autocorrelation coefficient and variance of EP signal; compared with autocorrelation coefficient, variance is more effective for prediction of the critical failure point of the rock specimens.