Objective To systematically evaluate the risk prediction models for neonatal early-onset sepsis (EOS), aiming to provide reference for the construction and optimization of models, as well as for clinical selection of appropriate prediction models. Methods PubMed, Web of Science, Embase, Cochrane Library, China National Know-ledge Infrastructure (CNKI), Wanfang Data, China Biology Medicine disc (CBM), and VIP databases were retrieved, and studies relevant to neonatal EOS risk prediction models were collected. The retrieval period was from the inception of the database to January 18, 2025. Two researchers independently screened literatures, extracted data, and evaluated the quality of the included literatures using PROBAST tool. Any disagreements were resolved through consultation with a third reviewer. Results A total of 14 literatures were included in analysis, containing 19 risk prediction models. The area under receiver operating characteristic (ROC) curve (AUC) of the included model ranged 0.71-0.999. The number of prediction factors ranged 3-21. Common prediction factors included young gestational age, low birth weight, 1-minute Apgar score, abnormal neonatal temperature, prolonged premature rupture of membranes, amniotic fluid turbidity, maternal Group B streptococcal infection, maternal chorioamnionitis, as well as elevated levels of procalcitonin and C-reactive protein in neonates. The risk of model overall bias was high, mainly due to insufficient number of outcome variable events in the analysis field, improper processing of missing data, screening of prediction factors based on univariate analysis, lacking model performance evaluation, and overfitting of model. Conclusion The neonatal EOS risk prediction model is still at the development stage. Although the current prediction models have better overall predictive performance, the overall quality needs to be improved. Future modeling can follow the PROBAST and TRIPOD specifications to reduce bias risk, explore the combination of multiple modeling methods, and focus on strengthening external validation and localized application to enhance the clinical applicability and promotion value of the model.