<p>Fatigue is a complex condition characterized by a decline in a person’s mental or physical performance. Methods to gauge fatigue include self-reported questionnaires, electroencephalography and camera-based technologies. However, these methods are typically restricted to laboratory settings, which limits their wider accessibility. Here we report a soft on-eyelid magnetoelastic sensor that can capture eye-blink parameters in real time and quantitatively decode fatigue levels. The sensor, which works in a self-powered manner, comprises a magnetomechanical coupling layer formed from a silicone rubber matrix embedded with micromagnets and a conductive gold coil patterned onto a thin thermoplastic elastomer layer. This design allows the conversion of eye movements into high-fidelity electrical signals. The sensor exhibits a Young’s modulus of 200 kPa, a stretchability of up to 530% and a pressure sensitivity of 0.2 µA kPa⁻¹. Its thin membrane structure adheres conformally to human upper eyelid tissue and maintains intimate contact during diverse eye movements. When combined with a one-dimensional convolutional neural network and data-processing techniques, the sensor can recognize subtle eye movements and categorize fatigue levels with an accuracy of 96.4% based on six eye-blink parameters.</p>