A method for computing probabilities of failure of structural componen
ts and systems is proposed. The approach is based on a radial-space di
vision technique in conjunction with automatic generation of unit cent
erline vectors. Based on this technique, the standard normal space is
divided into active and passive subdomains. The probability contributi
on of each active subdomain to the total probability of failure is est
imated by approximating the actual limit-state hypersurface with a hyp
ersphere segment centered on the actual hypersurface. The proposed met
hod can solve with relatively high accuracy both component and system
reliability problems involving nonlinearities and multiple extremum po
ints of the probability density on the limit-state hypersurface. A num
ber of numerical examples involving linear and nonlinear performance f
unctions of components and systems are presented. The results are in g
ood agreement with exact solutions, probability bounds, multinormal ap
proximations, and/or Monte Carlo simulations.