The invention discloses a power distribution network fault diagnosis method based on synchronous phasor measurement and a covariance matrix theory. The method comprises the following steps: monitoringrunning of a power system, regularly uploading amplitude values and phase information of a three-phase voltage, a current and a zero-sequence current, calculating to obtain amplitude values and phaseinformation of positive and negative zero-sequence components; using the amplitude values and phases of the real-time three-phase voltage and current as input features, forming a continuously updatedstate judgment matrix, while an element in the state judgment matrix generates saltation, considering that a fault occurs, and determining a fault occurrence moment; using amplitude values and phasesof the positive-sequence current and voltage as the input features, and forming a fault section judgement matrix, using amplitude values of the positive, negative and zero-sequence current and voltage under different types of the faults as the input features, forming a fault type judgement matrix; and through analyzing elements in the formed fault section judgement matrix and fault type judgementmatrix by using a covariance matrix algorithm, according to a relation between the features of the corresponding matrix, acquiring a fault section and type.