Reliability sensitivity algorithm is presented on thebasis of subset simulation and importance sampling due to the small failureprobability highly experienced in engineering. Firstly, a small failureprobability is expressed as a product of larger conditional failureprobabilities of some intermediate failure events. Secondly, the largerconditional failure probabilities can be estimated efficiently byconstructing the importance sampling density functions of the intermediatefailure events. Thirdly, the reliability sensitivity is transformed into thepartial derivatives of conditional failure probabilities with respect to thedistribution parameters of the basic variables in the paper. The estimationof the reliability sensitivity and its variance are then derived for thepresented algorithm. The results from several cases show that the presentmethod is efficient, precise and applicable to the structural system withsingle and multiple failure modes.