This paper deals with the problem of diagnosing the technical condition of a planetary gearbox operating at variable load. The severity of the subject and related difficulties were discussed. Theoretical basis of analysis of non-stationary signals (order analysis) and its use in signal resampling was also presented. The paper tests the functionality of the planetary gearbox diagnostics method. The Multilayer Perceptron Network was used to identify and classify the damage. The network’s learning vectors were built on the basis of order analysis results and measurements of the planetary gearbox load. The functionality of two-layer and three-layer unidirectional artificial neural network was also analysed for potential use in diagnosing the technical condition of planetary gears.