In many engineering applications, the probability distributions of some random variables are truncated; these truncated distributions are resulted from restricting the domain of other probability distributions. If the first order reliability method (FORM) is directly used, the truncated random variables will be transformed into unbounded standard normal distributions. This treatment may result in large errors in reliability analysis. In this work, we modify FORM so that the truncated random variables are transformed into truncated standard normal variables. After the first order approximation and variable transformation, saddlepoint approximation is then used to estimate the reliability. Without increasing the computational cost, the proposed method is generally more accurate than the original FORM for problems with truncated random variables.