To address the unmet need for a low-cost, low-power wearable solution for continuous cardiovascular health monitoring, we developed and validated an atrial fibrillation (AF) detection algorithm using clinical data collected with a microelectromechanical system (MEMS)-based pressure sensor. This sensor system, consisting of a circuit board, capacitive digitizer, and three MEMS elements, was specifically designed for early detection of AF—a common cardiac arrhythmia that requires frequent screening. The proposed algorithm extracts seven AF-related features, derived from autocorrelation analysis, interbeat interval (IBI) measurements, and differential IBI (dIBI) analysis, including a novel mean distance of points in the Poincaré plot (MDPP) feature. Clinical validation was conducted using data from 53 participants across three datasets: 13 healthy volunteers (wrist), 20 postcardiac surgery sinus rhythm (SR) patients (wrist), and 20 patients with AF (wrist and finger). Leave-one-out cross-validation showed that logistic regression achieved an area under the receiver operating characteristic curve (AUROC) of 93.0% using the full feature set. Performance remained stable across segment lengths ranging from 10 to 120 s, supporting the algorithm’s suitability for continuous monitoring. Consistent performance across seven different classifiers (average AUROC 92.1%) further demonstrated the clinical applicability and generalizability of the approach for wearable-based AF screening. To assess robustness against motion artifacts, we introduced five types of synthetic noise, with the algorithm maintaining strong AF detection performance under these conditions. Finally, a systematic evaluation of sensor waveform shape and signal strength across SR and AF at both the wrist and finger sites demonstrates the potential of the sensor system for wearable AF screening.