Abstract. The Norwegian company EWOS AS produces fish feed for the salmon farming industry, supplying approximately 300 customers spread along the coast of Norway. The feed is produced at three factory locations and distributed by a fleet of 10 dedicated vessels. The high seasonality of the demand and the large number of customers make the distribution planning a substantial challenge. EWOS handles it by operating a system of mostly fixed routes with decentralized planning at each factory. The distribution can be described as a multi-depot vehicle routing problem with time windows, multiple vehicle usage, inter-depot routes, heterogeneous fleet and a rolling horizon. The paper presents a mathematical model for this problem, which is solved by heuristics and meta heuristics. Based on detailed historical data collected by EWOS during the autumn of 2010, the model has proposed a dynamic set of routes with a significant reduction of travelled distance -close to 30% -and an increase of average vessel fill-rate -from 60% up to 95%. This implies a substantial fuel saving, with a positive environmental impact, and also a potential for downscaling the fleet, with additional considerable cost savings for the company.