The demand forecast for blood bags has the purpose of contributing to a more efficient management of blood centers, directing more effective collection campaigns and, consequently, facing the problems of stockouts that threaten the quality of the service provided. The problem, then, lies in the efficiency in the management of blood bags, since there is no targeting or forecast demand for these items in blood centers in the region. The proposed method is organized in five phases: (i) extraction of data sets through relational databases, (ii) organization of the database, (iii) verification of the consistency of the database, (iv) application of the method ARIMA for forecasting blood bags and (v) analyzing the performance of the forecasting model. Data sets were obtained with visualization of their respective time series. Subsequently, the data were treated for application of the ARIMA method afterwards, where discrepant data were identified in blood types AB +, AB-, B +, B- and A-. These blood types were treated using Shapiro-Wilk's normality tests and outliers were replaced by the mean. According to Adfuller's representation of autocorrelations, partial autocorrelations and stationarity tests, the parameters used by the forecast model were configured. An automatic algorithm (auto.ARIMA) was also implemented to detect the best parameters to be used. After separating the data sets in training and testing, the performance of the model for blood types was verified, and two of them (A + and O +) showed lower proportional errors since they have their highest blood bag values. The model also presented forecast indicators according to real demand, whether negative or positive. The blood type AB- obtained a greater error, making its use in practical cases not indicated. It was also found that the data sets have great randomness in the data, and the multivariate analysis of time series that cover festive periods, elective surgeries, among others, may be interesting.