This work presents different opposite learning strategies for Ant Knapsack, an ant based algorithm for the Multidimensional Knapsack Problem. We propose to include a previous opposite learning phase to Ant Knapsack, for discarding regions of the search space. This opposite knowledge is then used by Ant Knapsack for solving the original problem. The objective is to improve the search process of Ant Knapsack maintaining its original design. We present three strategies which differ on how the solutions can be constructed on a opposite way. The results obtained are promising and encourage to use this approach for solving other problems.