What I like about it is the fusion of algorithm and art. Of course there is much generative algorithmic visual art, but this work is rather distinguished in its particular fusion.
For instance, Box.Fitting.Img is both beautiful visually and, also, the work grows from an algorithm that one may easily infer from watching the piece. It starts with 5 boxes. The color of the boxes is determined by the color of the pixel of an underlying, invisible image. Though as the piece grows, one gets other indications of the underlying image. In any case, a box grows until it touches another box. Then it stops growing and other boxes start growing in the interstices remaining.
Very simple algorithm. Plain to see. But brilliantly so, really, and unusual in its visual results.
And much of his work is this way: the algorithms are evident if you watch closely. They are simple but often generative of unusual results. And his sense of color and shape is finely drawn. No clumsy grab bag goin on here. The sense of composition is fascinating. Composition within a pseudo-random generative process.
Another wonderful part of the work is that all the source code is available to view. It's all done in a language called Processing invented not too long ago by Casey Reas and Ben Fry. Java based. Some other strong work is emerging from this language such as Martin Wattenberg's, and Marek Walczak's "Thinking Machine" at http://www.turbulence.org/spotlight/thinking , which we have discussed a bit on the list.
In Tarbell's work we see the strong abstract quality of generative computer visuals. Strong abstract and dynamic properties/mode/process/aesthetic. Also we see the value of the source code to the artist-programmer community.
Another *very* strong site by Tarbell is http://levitated.net/daily .