The invention relates to a multiscale energy feature linear recognition method for a partial discharge ultrahigh frequency signal of a transformer, and belongs to the technical field of partial discharge mode recognition of a power transformer. The method comprises the following steps: S1, collecting the partial discharge ultrahigh frequency signal data; S2, decomposing the partial discharge ultrahigh frequency signal through a wavelet packet, and obtaining multiscale signals; S3, extracting multiscale energy parameters; S4, constructing a linear discriminant analysis method, and carrying out the recognition of the partial discharge ultrahigh frequency signal. The method is parallel to a conventional partial discharge signal statistical characteristic pattern of a power transformer, and the distinguishing characteristics lie in that the waveform features of the ultrahigh frequency signal are abundant; the multiscale energy features of a single extracted signal are linear; a linear classifier can be employed for effective recognition, thereby greatly improving the calculating speed of partial discharge signal recognition and effectively improving the efficiency of state monitoring and fault diagnosis of the power transformer.