This paper describes an estimation method of the structural failure probability based on an efficient Monte Carlo simulation, with which variance reduction methods of a directional importance sampling and a partition of the region are combined to improve the simulation efficiency. The structural failure probability is formulated by using a radial variable and a directional variable. Samples of the radial variable are generated from a truncated chi-square p.d.f. defined outside the β-sphere region. And instead of constructing a directional importance sampling p.d.f. beforehand, directional variable samples are determined from those generated by an importance sampling p.d.f. centered at the design points on the limit state surfaces in the rectangular coordinates and the probability volume contained in a hyperconical domain subtended by an infinitesimal increment at the respective determined directional variable is evaluated numerically and adopted it equivalently as a directional importance sampling probability density of the sampled direction. Numerical examples show that the proposed method gives accurate estimations of structural failure probabilities efficiently.