The invention discloses a clustering algorithm-based exceptional event analysis method for evaluating the whole state of an electric meter. The method comprises the steps of aiming at the abnormal users with overvoltage, overcurrent, wrong clock of the electric meter, undervoltage of the electric meter, overhigh temperature of the electric meter, opened cover of the electric meter, continuous exceeding on the upper limit of a load, voltage reverse phase sequence, current reverse phase sequence, inversed current, imbalanced voltage three phases, imbalanced current three phases and the like, the document information of the users are checked, and the users having documents with problems are filtered; for the users with the abnormity, the whole state of the electric meter is evaluated and analyzed by combining information such as marketing service metering fault, fault meter changing and information lack examining on site through the comprehensive analysis of the clustering algorithm according to multiple dimensions such as the manufacturer, the batch, the region, the type of the metering device, the voltage level, the user class and the line, a technical means and a reference basis are provided for whether an electric power company for on-site inspection needs to be performed and whether a shift plan needs to be made, and the method is accurate and has small errors.