To acquire a rational minimum cut-off time and the precision of lifetime prediction with respect to cut-off time for the accelerated aging test of LED lamps, fifth-order moving average error estimation is adopted in this paper. Eighteen LED lamps from the same batch are selected for two accelerated aging tests, with 10 samples at 80 ℃ and eight samples at 85 ℃. First, the accelerated lifetime of each lamp is acquired by exponential fitting of the lumen maintenances of the lamp for a certain cut-off time. With the acquired lifetimes of all lamps, the two-parameter Weibull distribution of the failure probability is obtained, and the medium lifetime is calculated. Then, the precision of the medium lifetime prediction for different cut-off times is obtained by moving average error estimation. It is shown that there exists a minimum cut-off time for the accelerated aging test, which can be determined by the variation of the moving average error versus the cut-off time. When the cut-off time is less than this value, the lifetime estimation is irrational. For a given cut-off time, the precision of lifetime prediction can be computed by average error evaluation, and the error of lifetime estimation decreases gradually as the cut-off time increases. The minimum cut-off time and medium lifetime of LED lamps are both sensitive to thermal stress. The minimum cut-off time is 1104 h with the lifetime esti-mation error of 1.15% for the test at 80 ℃, and 936 h with the lifetime estimation error of 1.24% for the test at 85 ℃. With the lifetime estimation error of about 0.46%, the median lifetimes are 7310 h and 4598 h for the tests at 80 ℃ and 85℃, respectively.