This research is aimed at overcoming the limited forage problem at marginal land that adversly affects the sustainability of forest. The research was done at Gerokgak District, Buleleng Regency, Bali Province from April 2010 until March 2011. Because the feed problem is complex and dynamic, systems dynamic modelling was used as the method. This method supported with participatory rural appraisal methods for system identification, mean absolute percentage error method for analysis validity of model, and sensitivity analysis to determine the scenarios of feed scenarios. Simulation results showed that the feed available on the actual conditions (in 2009) is only able to meet 64.5% of the total feed required, and potentially to 48.5% in the long term (year 2034). In that period, the feed supply of 52,629 tons / year increased to 74,886 tons/year while consumption increased from 81,403 tons/year to 155,834 tons/year, thus it potentially increases the damage of forest because the burden of forests as the source of feed would increase from 35.5% to 51.5%. To overcome these problems, five scenarios was developed to increase the feed supply. From the five alternative scenarios, the highest increase of feed supply successively given by Scenario 2 (planting field boundary with forage plants), followed by Scenario 1 (introducing the new technology in maize and rice cultivation), Scenario 5 (building feed storages), Scenario 4 (increasing the number of forest community groups from four to 14) and the last is the Scenario 3 (plantingcorn in the dry season by utilizing irrigation ponds) Those scenarios give the potential additional supply of feed 51.6%, 30.1%, 10.1 %, 3.1% and 0.1% respectively. Partially, no single scenario can provide sustainable feed supply; to do so the combination of at least two scenarios must be implemented.keywords: forage supply model, Bali cattle, marginal land, system dynamicRai