The invention discloses a power system adverse data identification method comprising the following steps: estimating system states so as to obtain a residual error vector and residual error sensitivity matrix; building an adverse data identification sparse optimization model; using an optimization method to solve the sparse optimization model so as to obtain a measurement error vector; comparing each element in the measurement error estimated vector with a preset threshold, and determining elements greater than the threshold as adverse data. The power system adverse data identification method abandons a conventional hypothesis examination probability theory concept, and innovatively develops the sparse optimization model, thus converting the adverse data identification problem into an optimization problem, and greatly improving practicality and reliability.