OBJECTIVES:  To construct a model in order to predict the hospitalization cost of patients based on demographic, clinical and care-associated characteristics METHODS: Based on the records of health care services for more than 3 million affiliates to a Colombian insurer, a base with 68,939 inpatient care services for 1 year was built. The analysis was developed using a regression model, where the dependent variable is the natural logarithm of the hospitalization cost per day and the independent variables were dichotomies. The multilevel model was adjusted using the smearing coefficient. We performed the necessary tests to validate the biggest possible explanatory capacity of the multilevel model: tests of individual statistical significance (t-test) for each variable and the level of explanation of this model compared to others (Chi2 test). RESULTS: A multilevel model with three levels was obtained: the characteristics of patients in the first level (age, gender, type of service and diagnosis of admission or discharge), previous comorbidity conditions in a second level and characteristics of Hospital development in a third level. All variables were statistically significant at a 95% confidence level. The multilevel model presented a higher level of explanation compared to the linear model, as indicated by the Chi2 test CONCLUSIONS: It was possible to construct an explanation model that allows to estimate the hospitalization cost in order to be able to improve the management and allocation of the resources.