Introduction: Current technologies to identify ventilations during out-of-hospital cardiac arrest (OHCA) are limited. Thoracic impedance (TI) recorded by defibrillator pads can be used to identify ventilations, but these are often obscured by artifacts due to chest compressions (CC) and electrode motion. We tested a novel automatic algorithm to detect ventilations in TI during continuous CC. Methods: We analyzed Philips MRx CPR processed data from Dallas - Fort Worth patients in the Pragmatic Airway Resuscitation Trial. We extracted one-minute TI segments during continuous CC with concurrent capnogram recordings. We applied a novel algorithm to identify ventilation in the TI signal, which combined a) adaptive signal processing to remove CC artifacts, b) a recurrent neural network to discriminate TI fluctuations as ventilations, and c) a quality control stage to anticipate TI segments in which ventilation detection could be defective. We evaluated the performance of the algorithm in detecting ground truth capnogram ventilations in terms of sensitivity (SE), proportion of correctly identified ventilations, and Positive Predictive Value (PPV), proportion of detections corresponding to actual ventilations. Results: We extracted 2,551 one-minute TI segments from 367 OHCA patients; median of 6 (IQR 3-10) minutes per patient. Median patient-wise SE and PPV were 86.5 (IQR 71.6 - 95.1) % and 85.4 (68.3 - 94.7) %, respectively. The quality control was successful at discriminating TI segments by ventilation detection performance (Figure). For the 75% and 50% of data with highest quality estimates, SE improved to 90.6 (76.8 - 96.7) % and 95.5 (87.1 - 100.0) %, respectively, and PPV improved to 89.5 (75.0 - 96.4) % and 94.9 (87.3 - 99.8) %, respectively. Conclusions: We demonstrated the accuracy of an algorithm to detect ventilations using TI during continuous CC. Ventilation detection using TI during continuous CC is feasible.