We prove an upper bound on the Shannon capacity of a graph via a linear programming variation. We also show that our bound can be better than Lovász theta number and Haemers minimum rank bound.