This study describes an “important quadrants sampling method” for the simulation-based estimation of structural failure probability． In the first stage, generating real samples iteratively in arbitrary quadrants of the basic variable space and applying the previously proposed “inter-quadrant relational expression,” the coordinates of the real sample points generated are transformed to those of pseudo sample points over the all quadrants． Among the all quadrants, several quadrants whose coordinates of pseudo sample points judged to contribute to the estimation of structural failure probability, are determined as the “important quadrants．” In the second stage, the estimation of the failure probability of structural system based on the directional simulation is executed by using the pseudo samples transformed in the important quadrants transformed from real samples generated in the arbitrary quadrants． Numerical examples show that the proposed method gives effectively accurate estimations of structural failure probabilities.