This paper presents a cost-effective probabilistic approach to be used in engineering applications. The proposed approach consists of an improved Kriging-based method aiming at reducing to a minimum the number of evaluations of the true performance function when computing a failure probability. It is a kind of variant of the classical Active learning method combining Kriging and Monte Carlo Simulation (AK-MCS) developed by Echard et al. (2011), where some improvements are introduced to enhance the learning process. Some illustrative and practical examples are presented and discussed. The proposed approach has shown a great efficiency as compared to the classical AK-MCS approach.