There are many methods to extract fetal electrocardiogram (FECG) from multi-channel abdominal signals. The blind source separation is widely concerned, which is based on the assumption of statistical independence and non-Gaussianity among signals. However, due to the influence of maternal ECG (MECG) signals and a variety of complex bioelectrical noises mixed in the abdominal ECG signals, as well as the unreasonable arrangement of the abdominal electrodes, the extraction results of FECG by the traditional blind source separation algorithm are often unsatisfactory. In this paper, an EMD-QPCE method was proposed for the extraction of multi-channel FECG based on the empirical mode decomposition (EMD) and quasi-periodic component extraction (QPCE). Firstly, the abdominal mixed ECG signals of each channel were decomposed into a series of intrinsic mode functions (IMF) by EMD, and the components of MECG signals in the IMFs were eliminated to enhance the FECG information. Then, the IMFs of the same order in each channel signals were combined and the IMFs with the FECG information were extracted by QPCE. Finally, an FECG was reconstructed from the IMFs containing the FECG information. Experiments were carried out on real ECG signals in DaISy, ADFECGDB and Challenge 2013 databases. The results showed that compared with the traditional ICA, PCA and QPCE methods, the EMD-QPCE method proposed in this paper greatly improved the success rate and the quality for the FECG extraction. Within the experimental scope, Sensitivity of extraction (Se) for FECG of the proposed method was above 92.3%. Positive predictive value (PPV) was greater than 98.8%. Accuracy index F1 was greater than 95.4%. And average fetal heart rate error was less than 0.595%. It showed a very good accuracy, which verified the effectiveness and superiority of the proposed method.