An interharmonic spectrum estimating method combining the features of parametric method and nonparametric method is presented to detect the interharmonic in power system. The signals from power network are transformed to spatial signal based on the spatial signal analysis， and the number of interharmonic sources and their frequencies are accurately estimated with the help of Gerschgorin's disks combining root multiple signal classification（Root-MUSIC）. The rest amplitudes and phase angles are then given by a robust support vector regression machine（SVRM）. After frequency information having been obtained， the frequency range determined by SVRM is reduced to known frequency dots and calculation is cut to minimum to avoid synchronous sampling and over-long sampling data. The digital simulation and actual inspection prove the method is effective and applicable.