In the past, winter maintenance in Flanders started in October with large fixed stocks. In total, approximately 38.000 tons of salt were stored in several depots at the beginning of the winter season. The stocks were built up during summertime at lower prices (up to 5 per cent cheaper) and, correspond to the demand for salt during an average winter period. In function of the salt usage and the contractual delivery times new salt was ordered during wintertime at winter prices. Only near the end of winter, an attempt was made to reduce the stocks. This policy certainly has some advantages: it is easy to implement and allows to operate relatively independent of the salt suppliers. However, its main drawback is that you may end up with large amounts of unused salt at the end of wintertime. These stocks have to be kept until the next winter period and will incur large holding costs. To cope with this problem a new stock model based on an (R,S)-inventory policy is developed. The idea is to match the salt inventories in the depots closer to the actual demand for salt during the various winter months and so, in combination with shorter delivery times and better exchange of stock information between the salt suppliers and the road administration, reduce the stocks. In an (R,S)-inventory policy, as soon as the stock reaches a minimal level R (reorder point), an order is placed to bring the stock to a maximum level S (stock level or target level). Both minimum and maximum stock levels are not fixed during the whole winter period but will vary each month and for each district. The model parameters R and S are determined for a predefined service level and, in function of the historical spread and weather data. Two approximations are used. The first is based on a multi-linear regression (the weather type predictions made in the past are matched with the salt usage in the past corresponding to these weather types), the second approximation is based on pure historical data of salt amounts spread. The models allow decision support to know when and how much salt must be ordered. The first model is more accurate, the second one more easy to use. Both models still need further fine-tuning (weather data, based on daily data) but have proved to be useful. Also relevant financial gains seem to be possible. Each district in the road administration will now, after the implementation of the model for the next winter and after feeding the model with the district specific data (amount of square meters to be spread, climatologic zone, safety parameters) manage his own salt stocks based on a uniform decision making process. The model must be fed with new (recent) data to become more and more accurate.