Objective To construct the prediction model of preeclampsia (PE) in early pregnancy by analyzing the general data, risk factors and routine laboratory indicators of pregnant women for 6-10 weeks, and compare the prediction ability of Logistic regression model and eXtreme Gradient Boosting (XGBoost) model. Methods The general data, risk factors of PE and 27 routine laboratory indicators such as blood lipids, liver and kidney function, blood coagulation, blood cell count and etc. during 6 and 10 weeks of gestation from 925 PE patients and 7 613 normal controls in Peking University Third Hospital from January 2015 to August 2020 were retrospectively analyzed by the statistical methods such as Mann-Whitney U test, Logistic regression and XGBoost. And then, the prediction model was constructed, the receiver operating characteristic curve (ROC) was drawn, and the area under the ROC curve (AUCROC), sensitivity and specificity were calculated. The feature importance bar chart was drawn with XGBoost. Results There were significant differences in the proportions of diabetes, SLE, antiphospholipid sysdrome, renal diseases, whether there was a history of eclampsia or PE and primipara between the two groups (P＜0.05). Among the 27 routine laboratory indicators, there were significant differences in all other indicators except Plt/Lym between the two groups (P＜0.05). The AUC (0.621, 95%CI=0.601-0.640, sensitivity=34.8%, specificity=81.5%) of Logistic regression model established based on 7 risk factors was lower than that (AUC=0.752, 95%CI=0.735-0.769, sensitivity=64.2%, specificity=76.0%) of combined Logistic model based on 6 risk factors and 14 laboratory indicators and that (AUC=0.867, 95%CI=0.839-0.896, sensitivity=73.0%, specificity=82.3%) of XGBoost model, indicating that XGBoost model had the best ability to predict the early onset of PE. The top three indicators of importance screened by XGBoost were TG, Lp(a) and C1q. Conclusion The efficacy of using risk factors combined with routine laboratory indicators to predict PE in early pregnancy is better than that of using clinical risk factors alone, and XGBoost model in early predicting the risk of PE is better than the Logistic regression model. TG, Lp(a) and C1q are important variables for early prediction of PE.