The invention provides a feature extraction method for a turnout action current curve, and the method comprises the following steps: sampling a turnout action current, and building the turnout action current curve; marking the last current data in the turnout action current curve as a conversion end point D; searching the current data with the maximum current amplitude, and marking the current data with the maximum current amplitude as a turnout start point A; sequentially calculating a j-th current data point between the turnout start point A and the conversion end point D and the mean value (shown in the description) of the following (n-1) current data points and a variance K, comparing the variance K with a preset threshold x, setting the value of the variance K as one when K is not greater than x, and setting the variance K as zero when K is greater than x; building a variance curve graph, marking the current data corresponding to a first rising edge start point in the variance curve graph as a turnout action current start point B according to the correlation among the variance K, the variance K and the variance K, and marking the current data corresponding to a last falling edge end point as a turnout action current end point C.