In the online version of Self-Organizing Maps, the results obtained from different instances of the algorithm can be rather different. In this paper, we explore a novel approach which aggregates several results of the SOM algorithm to increase their quality and reduce the variability of the results. This approach uses the variability of the algorithm that is due to different initialization states. We use simulations to show that our result is efficient to improve the performance of a single SOM algorithm and to decrease the variability of the final solution. Comparison with existing methods for bagging SOMs also show competitive results.