We introduce FEEL, a benchmarking study evaluating 19 emotion datasets based on physiological signals, uncovering key insights into their generalizability and cross-dataset transferability.