I only work with DNA, and as far as I know the NJ method is normally based on genetic distances, so for me seems that negative branch lenghts can be related to some error in the analysis, or in the data entry or in the calculation software. I suggest that you try other methods besides NJ to see if they gave the same results.

questions:

1. How is it possible to infer about phylogeny with a phenetic method (NJ method) ?

By this way, you'll obtain similitude's reconstruction! but not phylogenetic tree

2. wouldn't be more convenient to try an other tool ?

I understand very well that could be useful in a first approach, but to propose "genuine" phylogenetical hypotheses, it could be more interesting to use a parsimony approach.

Hey Thanks bruno and Priyanka

Any suggestions about freely available softwares for this?

Use MEGA, it's a free software where you can try different methods.

Thanks Boopathi

UPGMA is neither a phylogenetical approach, but a global similitude one !

and K2P is a model of ponderation of the transitions. I think they are not proper tools for the here described problem ! Use Paup. You cannot compare results with negative values, as the tools are fundamentaly different !