Objective To identify risk factors for respiratory failure induced by severe pneumonia in children and to develop a Nomogram prediction model. Methods A total of 223 children with severe pneumonia admitted in Xuchang Central Hospital from January 2021 to December 2023 were retrospectively enrolled. Clinical, laboratory and radiological data were collected within 24 h of admission. Cases were randomly divided into a training set (n＝156) and a validation set (n＝67) at a 7 vs. 3 ratio via computerized random sampling in SPSS 26.0. Candidate variables with P＜0.05 in univariate analysis were entered into multivariate Logistic regression (Enter method); variance inflation factors (VIF) were calculated to assess multicollinearity. Finally, the variables with a P value less than 0.05 were retained to construct the Nomogram prediction model. Discrimination, calibration and clinical utility were evaluated using receiver operator characteristic (ROC) curve, Hosmer-Lemeshow test, calibration curves and decision curve analysis (DCA). Internal validation was performed by 1 000 bootstrap resamples, yielding optimism-corrected area under the curve (AUC) and calibration slope. Results The incidence of respiratory failure in the training set was 39.74% (62/156), and in the validation set it was 38.81% (26/67). There was no statistically significant difference in the incidence between the two groups (χ2＝0.018, P＝0.893). Multivariate Logistic regression identified 6 independent factors: hypoxemia, acute physiology and chronic health evaluation Ⅱ score ≥15 points, age ＜3 years, multi-lobar infiltration, elevated interleukin-6 (IL-6), and procalcitonin (PCT) ＞0.5 ng/ml. All variables had a VIF of ＜5, and no significant multicollinearity was observed. The AUC of the training set was 0.827 (95%CI: 0.756 ? 0.892), with a sensitivity of 79.03% and a specificity of 73.40% at the optimal cut-off point. The AUC of the validation set was 0.749 (95%CI: 0.625 ? 0.861). The Hosmer- Lemeshow test showed that the χ2 value for the training set was 5.898, P＝0.659, and for the validation set it was 5.743, P＝0.676, indicating acceptable model calibration. The bias-corrected AUC obtained from 1, 000 Bootstrap resampling was 0.794, the calibration slope was 0.831 (95%CI: 0.558 ? 1.165), and the intercept was ?0.102 (95%CI: ?0.525 ? 0.378). The DCA results showed that the net benefit of the model was higher than that of the full intervention and no-intervention schemes within the threshold probability range of approximately 2% to 88%. Conclusion The Nomogram model constructed based on six early available indicators upon admission demonstrated acceptable discrimination and calibration in the dataset of this center, and can be used as a reference for early risk stratification of respiratory failure. Given that the current validation of this model is limited to internal validation and constrained by the single-center, retrospective design, and the sample size of the validation set, the extrapolability of the model still needs to be verified through multi-center, prospective studies.