Abstract: The Number Partitioning Problems（NPP），arising from the bin packing problem and line balancing problem in daily life，pose considerate challenge to both exact and heuristic solution methods.This paper introduces an improved estimation of distribution algorithm to the solution of NPP，adopting real coding，matrix-based storage of probability vectors and weight coordination method.Comparison is made between the standard Differencing Method（DM） and the improved EDA.Results show that the improved EDA has great advantage over DM in solving NPP of less than 25 dimensions.The algorithm is extended to higher dimension and multi-partitioning problems as a further exploration in its scalability.Some computational experience is presented here.