With the development of an implantable radio transmitter system, direct measurement of cardiac autonomic nervous activities (CANAs) became possible for ambulatory animals for a couple of months. However, measured CANAs include not only CANA but also cardiac electric activity (CEA) that can affect the quantification of CANAs. In this study, we propose a novel CEA removal method using moving standard deviation and cubic smoothing spline. This method consisted of two steps of detecting CEA segments and eliminating CEAs in detected segments. Using implanted devices, we recorded stellate ganglion nerve activity (SGNA), vagal nerve activity (VNA) and superior left ganglionated plexi nerve activity (SLGPNA) directly from four ambulatory dogs. The CEA-removal performance of the proposed method was evaluated and compared with commonly used high-pass filtration (HPF) for various heart rates and CANA amplitudes. Results tested with simulated CEA and simulated true CANA revealed stable and excellent performance of the suggested method compared to the HPF method. The averaged relative error percentages of the proposed method were less than 0.67%, 0.65% and 1.76% for SGNA, VNA and SLGPNA, respectively.