Owing to better-defined edges and enhanced contrast in multi-order harmonic imaging, the ultrasonic technique provides the improved diagnosis performance for small lesions in clinics. The separation based on filtering is a most frequently used method for extracting multi-order harmonics from ultrasonic radio frequency (RF) echo signals. However, the cutoff frequency, order, and type of a filter have a great influence on the precision of harmonic extraction. In the present study, an adaptive method based on the empirical wavelet transform (EWT) algorithm is proposed to adaptively separate multi-order harmonics separation from ultrasonic echo signals. First, a set of empirical mode functions (EMFs) is obtained from the ultrasonic RF echo signals by the EWT algorithm. Then, each EMF is classified into the corresponding fundamental and harmonics categories according to the boundary of the EMF. Finally, the fundamental and different orders of harmonic signals are obtained by accumulating separately these categories. In the in-vitro experiments, the proposed method is used to separate the ultrasonic RF echo signals from isolated pig kidneys and compared with the results of filtering method. The image contrasts, contrast-to-noise ratios (CNRs), and tissue-to-clutter ratios (TCRs) of the second- and third-harmonics based on the proposed method are increased by 7.69 % and 12.32 %, 8.94 %, and 11.92 %, as well as 13.52 %, and 17.07 %, respectively. In conclusion, the EWT method is superior to the filtering method in harmonic separation owing to the better adaptive characteristic. This method efficiently performs the improved separation of multi-order harmonics for B-mode imaging and can be applied in practice to obtain more accurate diagnostic information.