Predictive models for preterm infant mortality have been developed internationally, albeit not valid for all populations. This study aimed to develop and validate different mortality predictive models, using Spanish data, to be applicable to centers with similar morbidity and mortality. Methods Infants born alive, admitted to NICU (BW<1500 g or GA<30 w), and registered in the SEN1500 database, were included. There were two time periods; development of the predictive models (2009-2012) and validation (2013-2015). Three models were produced; prenatal (1), first 24 hours of life (2), and whilst admitted (3). For the statistical analysis, hospital mortality was the dependent variable. Significant variables were used in multivariable regression models. Specificity, sensitivity, accuracy, and area under the curve (AUC), for all models, were calculated. Results Out of 14953 included newborns, 2015 died; 373 (18.5%) in their first 24 hours, 1315 (65.3%) during the first month, and 327 (16.2%) thereafter, before discharge. In the development stage, mortality prediction AUC was 0.834 (95% CI: 0.822-0.846) (p<0.001) in model 1 and 0.872 (95% CI: 0.860-0.884) (p<0.001) in model 2. Model 3's AUC was 0.989 (95% CI: 0.983-0.996) (p<0.001) and 0.942 (95% CI: 0.929-0.956) (p<0.001) during the 0-30 and >30 days of life, respectively. During validation, models 1 and 2 showed moderate concordance, whilst that of model 3 was good. Conclusion Using dynamic models to predict individual mortality can improve outcome estimations. Development of models in the prenatal period, first 24 hours, and during hospital admission, cover key stages of mortality prediction in preterm infants.