For identification and decomposition of domestic appliance load, a new feature extraction method is proposed in this paper. First, we adopt wavelet filtering processing on the bus current and remove the background current from the load transient current by periodic difference. Then S-transform is used for the transient current after load switching to obtain an amplitude harmonic matrix. The matrix is dimensionally reduced and the corresponding features are extracted in the row and column directions by the bidirectional two-dimensional principal component analysis. Finally, the feature data is classified by the support vector machine. The experimental results indicate the proposed feature extraction method has an average recognition accuracy of 99.24% and a maximum recognition accuracy of 100% on six kinds of appliance from the BULED dataset. Especially, the proposed method can provide better performance on differentiating the loads with similar electrical characteristics.