Due to the tough competitions that exist today, most pharmaceutical distribution companies are in a continuous effort to increase their profits and reduce their costs. Actually, both shortage and surplus of goods can lead to loss of income for these companies. One of the problems in pharmaceutical distribution organizations which deal with public health and pharmaceutical products is how to control inventory levels by means of accurate sales prediction in order to prevent costs of excessive inventory also prevent losing their customers because of drug shortage. Accurate sales prediction is certainly a valuable management tool to meet the mentioned goals since this leads to improved customer service, also, reduced lost sales and costs. However, most pharmaceutical distribution companies in Iran are still using heuristic or traditional statistical techniques to make sales prediction for their products. Thus, the purpose of this research is to apply an innovative and reliable sales prediction method for pharmaceutical distribution companies.To make sales prediction for a pharmaceutical distribution company, we needed to have past sales records of each drug. Accordingly, we gathered sales data of three years from Pakhsh Hejrat Co. which is one of the leading pharmaceutical distributors in Iran. We chose neural networks as our basic tools for sales prediction since most traditional methods like ARIMA are incapable of modeling nonlinearities that exist in most real data; also, they need forecaster’s supervision for the parameter estimation phase. In fact, neural networks are versatile tools for sale prediction since estimation with neural networks can be automatized, and they have proved very effective in order to make prediction by handling non-linear input and output variables Due to the fact that we did not have enough past sales records of drug items, we came up to a new idea of grouping drugs to find group members and make use of co-members’ sales data for each other. Thus, we did a comprehensive network based analysis in order to find clique-sets and group members. Afterwards, we built sales forecasting models with three different approaches: a) ARIMA methodology for time series forecasting, b) Hybrid neural network approach for time series forecasting by means of each drug’s past recodes, and c) Hybrid neural network approach for time series forecasting by means of each drug’s past records and its group members’ past records. Our evaluations and results indicated that our new methodology (number 3 above) was the best methodology, and the weakest one was ARIMA model.