This research addresses the challenge of developing a universal deepfake detector that can effectively identify unseen deepfake images despite limited training data. Existing frequency-based paradigms rely on frequency-level artifacts introduced during the up-sampling in GAN pipelines, but these detectors often overfit to specific artifacts present in the training data, leading to suboptimal performance on unseen sources. To address this issue, the authors introduce FreqNet, a novel frequency-aware approach centered around frequency domain learning. FreqNet forces the detector to focus on high-frequency information, exploiting high-frequency representations across spatial and channel dimensions. Additionally, a frequency domain learning module is incorporated to learn source-agnostic features by applying convolutional layers to both the phase and amplitude spectra of Fast Fourier Transform (FFT) and Inverse Fast Fourier Transform (iFFT). Extensive experiments involving 17 GANs demonstrate the effectiveness of FreqNet, showing state-of-the-art performance (+9.8%) while requiring fewer parameters. The code for FreqNet is available at <https://github.com/chuangchuangtan/FreqNet-DeepfakeDetection>.This research addresses the challenge of developing a universal deepfake detector that can effectively identify unseen deepfake images despite limited training data. Existing frequency-based paradigms rely on frequency-level artifacts introduced during the up-sampling in GAN pipelines, but these detectors often overfit to specific artifacts present in the training data, leading to suboptimal performance on unseen sources. To address this issue, the authors introduce FreqNet, a novel frequency-aware approach centered around frequency domain learning. FreqNet forces the detector to focus on high-frequency information, exploiting high-frequency representations across spatial and channel dimensions. Additionally, a frequency domain learning module is incorporated to learn source-agnostic features by applying convolutional layers to both the phase and amplitude spectra of Fast Fourier Transform (FFT) and Inverse Fast Fourier Transform (iFFT). Extensive experiments involving 17 GANs demonstrate the effectiveness of FreqNet, showing state-of-the-art performance (+9.8%) while requiring fewer parameters. The code for FreqNet is available at <https://github.com/chuangchuangtan/FreqNet-DeepfakeDetection>.