Abstract The system reliability is significant importance because it is involved into most stages of the whole life-cycle of product. According to the existing literature, MTBF is the most frequently used index to indicate the system reliability. However, in engineering calculation, early failure data are always involved directly into MTBF’s estimation, which results in that the estimated MTBF is always smaller than the actual MTBF, especially for the small samples. In order to alleviate this issue, the paper presents three integration algorithms to reconstruct PDF based on Weibull distribution, by reallocating the early failure data. Then relevant prediction expressions are derived by the reconstructed PDFs. Eventually, three new prediction methods are illustrated by a real example. As a result, the prediction value of Method 3 is very close to the actual observed value, moreover the prediction deviation rate is only 5.1%, which proves the proposed algorithm is reasonable. Finally the proposed methods are suggested applying to other distribution patterns.