This paper presents a novel approach for the identification of continuous-time systems directly from sampled I/O data based on trial iterations. The method achieves identification through iterative learning control (ILC) concepts in the presence of heavy measurement noise. The robustness against measurement noise is achieved through 1) projection of continuous-time I/O signals onto a finite dimensional parameter space and 2) Kalman filter type noise reduction. In addition, an alternative simpler method is given with some robustness analysis. The effectiveness of the method is demonstrated through numerical examples for a nonminimum phase plant.