Kmeans modification

i need an approach where some new idea is used for the modification of kmeans clustering algo that gives the better result than the existing kmeans...plz help me,,this is my project work.

[Walter Roberson]

Did you want any random new idea that _might_ give better results, or did you something more specific in mind?

[Geoff]

Do you know the number of clusters in advance, or are you searching for an 'optimal' number of clusters?