As the theoretical bases of accelerated life test for electronic components, failure distribution in stress g(stress) is deduced from life-time distribution f(t) by time-stress transformation assuming the Arrhenius model, exp(¸-Bb)t=constant, b=1/T for thermal stress T and the inverse power model, sat=constant for other stress s. When f(t) are assumed to follow lognormal distribution with parameters (μlnt,σlnt) and Weibull distribution with parameter (m,η), g(b) are normal distribution with parameters (μb,σb) and largest extreme distribution with parameters (mB,C/B) respectively and also g(s) are lognormal distribution with parameters (μlns,σlns) and Weibull distribution with parameters (ma,ηs) respectively. By making use of the relationships between parameters of f(t) and g(stress) such as B=σlnt/σb, B=mB/m and a=σlnt/σlns, a=ma/m, the acceleration constants B and a are effectively estimated from data of constant stress life test and step stress test. [In Japanese.]