Accurate diagnosis of the composite oil-paper insulation state of power transformers is of great guiding significance for the safe and stable operation of power systems and the operation and maintenance of equipment itself. In this paper, an evaluation method based on grey relational analysis (GRA) and a clustering cloud model is proposed to solve the problem of inaccurate evaluation caused by few characteristic quantities of the dielectric response of oil-paper insulation and failure to consider the randomness of the system. First, based on the recovery voltage method and the extended Debye model, five relevant features are extracted to establish the oil-paper insulation state evaluation system. Second, in view of the sensitivity differences of multiple feature quantities in the reactive insulation state, a combination weighting method combining GRA and an improved analytic hierarchy process is used to avoid data information loss and make the weight allocation more reasonable. Finally, it uses the atomization characteristics of the cloud model to reflect the randomness of the data, and comprehensively considers the randomness and fuzziness of the classification boundary of the evaluation index grade. After that a clustering cloud model membership selector is constructed. The validation of measured data from multiple transformers with different furfural content shows that the evaluation method can not only accurately reflect the actual insulation status of the transformer, but also reflect its deterioration trend, providing a reference basis for the formulation of maintenance strategies.