Load balancing is an important factor in a grid system to improve the global throughput of grid resources. Load balancing using genetic algorithm in grid systems improves the response time and the resource utilization and very effective in term of scalability. The proposed algorithm uses the GA to determine the weights of a multilayer feed-forward network with back-propagation learning. Conventional back-propagation networks make use of gradient descent learning to obtain their weights. However, there remains the problem of the network getting stuck in local minima. This problem is solved by combining the back-propagation with genetic algorithm.