Abstract: Scientific and accurate assessment of direct post-earthquake economic losses is important for reducing post-earthquake economic losses and developing emergency rescue plans quickly. In this paper, five factors were selected as assessment indexes: earthquake magnitude, epicenter intensity, the difference between epicenter intensity and seismic intensity, per capita GDP, and population density. Thirty-seven historical earthquake cases from 1996 to 2014 were chosen as research samples to construct the PSO-ELM post-earthquake direct economic loss assessment model. The experiments showed that the average error rate of the PSO-ELM model prediction was 13.89%, which reduced by 14.61% compared with the unoptimized ELM and 36.79% compared with the traditional BP network model. By analyzing the sensitivity of each influencing factor, it was concluded that per capita GDP, population density, and epicenter intensity had the most significant influence and should be focused on during post-earthquake direct economic loss assessment. In summary, the PSO-ELM model has the advantages of high accuracy and strong generalization ability and can be promoted and used in practical cases.