Heart defects are the leading cause of birth defect-related deaths. Monitoring fetal electrocardiogram (FECG) is very important for early detection of heart defects and abnormal FHR patterns. The detection of QRS complexes in FECG signals has a notable role in determining benchmarks correlated with fetal health e.g. fetal heart rate, intervals between each heartbeat, identification of congenital heart diseases, distress, Hypoxia etc. In this study a novel and automated approach based on deep learning methods has been introduced through applying which we’re be capable of detecting fetal QRS complexes from FECG signals. Data used in this experiment are collected from set-a of PhysioNet/computing in the cardiology challenge database (NI-FECGDB). This study proposes a 1-D Convolutional neural network architecture. The architecture of neural network consists of 5 convolutional layers, 7 batch normalization layers, 3 dropout layers and 3 dense layers. First step is consisted of preprocessing the data. In this step the data is being prepared to be used by the suggested approach through changing scale input signal, data augmentation and also building annotations that are applicable to the suggested network. Next step is consisted of application of the suggested algorithm. The performance of suggested method is the evaluated using evaluation criteria such as accuracy, mean error square, F1-score, sensitivity, specificity and precision, and following that the calculated results has been compared to the accumulated results of other studies done on NI-FECGDB database. In this study several methods have been used for evaluating the suggested neural network. In the best method accuracy, sensitivity, specificity and precision of the suggested method in detection of fetal QRS complex is 96.79%, 97.91%, 92.79% and 97.88% respectively. It is noteworthy that some of innovations of the suggested method are the capability of training the suggested network with only 20 of AECG signals and not removing maternal ECG from AECG signals.