A novel technique based on an improved growing hierarchical self-organizing maps（GHSOM） neural network for intrusion detection was presented.The improved GHSOM could deal with a metric incorporating both numerical and symbolic data

and then improved efficiency of intrusion detection.The validities and feasibilities of the improved GHSOM were confirmed through experiments on KDD Cup 99 datasets and simulated experiment datasets.The experi-ment results showes that the detection rate has been increased by employing the improved GHSOM.