Objective: Real-time detection of chest compressions during resuscitation can provide essential feedback to rescuers about CPR quality. Accelerometer technology can accurately detect compressions, but requires added equipment and expense. Conversely, the transthoracic chest impedance signal is recorded by nearly all defibrillators, and offers a more available means to characterize chest compressions. This work aims to develop and test an algorithm to accurately detect chest compressions using frequency characteristics of the impedance signal. Methods: To train the chest compression detection algorithm, 6-second impedance clips were taken from a convenience sample of 82 out-of-hospital cardiac arrest patients. Data from a separate group of 173 patients were used as a validation set to determine performance, using concurrently collected accelerometer data as the reference for true chest compressions. Clips were collected randomly during either compressions or pauses. Clips with transitions between compressions and pauses were excluded. The detection algorithm used bandpass filtering to eliminate ventilations and artifact outside normal chest compression frequencies. Spectral peak analysis between 1-4 Hz was employed to eliminate harmonics of the fundamental chest compression frequency and detect true compressions. Results: Of 1258 6-second clips from the 173 patients in the validation dataset (846 with chest compressions and 412 without), the sensitivity and specificity of the impedance-based compression detection algorithm were 98.7% [97.7, 99.4] and 96.4% [94.1, 98.0], respectively. Conclusion: An impedance-based algorithm can detect chest compressions with high accuracy using frequency analysis, and provides a potentially valuable tool for assessing resuscitation quality.