<p>Recently, the reliability analysis methods combining active learning Kriging strategies and Monte Carlo simulation (AK-MCS) are increasingly popular. Among them, active learning strategies based on the error analysis of the predicted failure probability are relatively efficient. However, most of the strategies fail to comprehensively consider the effect of updating the Kriging model on the overall prediction error. In this paper, a new active learning Kriging method, which is based on the quantitative analysis of the prediction error, is proposed for structural reliability. First, based on the statistical properties of the Kriging model, the rigorous error analysis of AK-MCS for predicting the failure probability is derived; then, the effect of updating the Kriging model with new training samples on the error of the predicted failure probability is analyzed. To effectively reduce the prediction error of the failure probability, a learning function combined with the pseudo Kriging strategy is proposed. Additionally, an error-based stopping criterion matching the learning function is developed. By combining the error-based learning function and convergence criterion, a new active learning Kriging method for structural reliability is proposed. Finally, four examples are used to verify the effectiveness of the proposed method. The results demonstrate the high efficiency of the proposed method in assessing structural reliability.</p>