dear Paul,
there are several wonderful packages to deal with DWI data. It depends a bit also on what you want to do, i.e. streamline vs. probabilistic tractography, or quantification of FA differences, for instance. I think that both FSL and Trackvis are very good starting points. 

In FSL (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/) you can do voxelwise analysis of FA differences (with TBSS) as well as tensor fitting and probabilistic tractography. A good point to start is their fslcourse (check out slides, tutorials and data from the previous editions), where you can find also specific tutorials for the DWI-related part: http://fsl.fmrib.ox.ac.uk/fslcourse/
N.B. for FSL I think it's better if you are using either an *NIX or Mac system, but you can also run it inside a virtual machine (e.g. Neurodebian)

In Trackvis (http://www.trackvis.org/) (together with the diffusion toolkit) you can do tensor estimation (and much more, including Q-ball and DSI), as well as several flavours of streamline tracking. It's very good if you're a GUI guy, although you can also take advantage of the very useful online commands

These are only a couple of them, but there are dozens of them. For instance, another wonderful package (with a steeper learning curve IMHO) is the Camino diffusion toolkit (http://cmic.cs.ucl.ac.uk/camino/), as well as all the MedInria suite (http://med.inria.fr/)

all the best!
enjoy DWI!

leonardo