The invention discloses a power grid abnormal state detecting method based on a maximum feature value of a sample covariance matrix. The power grid abnormal state detecting method based on the maximumfeature value of the sample covariance matrix comprises the following steps: step 1, constructing a data source matrix Xs; step 2, acquiring a sliding window matrix X; step 3, standardizing the sliding window matrix X; step 4, acquiring a sample covariance matrix S; step 5, solving a maximum feature value as shown in specification of the sample covariance matrix; step 6, power grid state abnormalout-of-limit judgment: judging whether the maximum feature value as shown in specification is greater than a threshold value as shown in specification, if the maximum feature value is greater than the threshold value, determining that the state of a power grid is abnormal, and giving an alarm, and if the maximum feature value is not greater than the threshold value, determining that the abnormalstate does not exist at present, and returning to step 2 to continue carrying out a state abnormity detecting flow. The potential invalidation problem caused when a traditional mean spectral radius method detects the abnormal state of the power grid in a low signal-to-noise ratio scene is solved, meanwhile, calculation consumed time of a traditional method for detecting the abnormal state of the power grid on the basis of a random matrix theory is saved by simplifying a calculating link, and the calculating efficiency is improved remarkably.