Granted, every model is wrong. No data drawn from the real world is truly Poisson, but it's an approximation to the distribution of some datasets, and I believe mine adequately conform. It's even going to be hard to test for conformity to a Poisson distribution with such a small sample, isn't it? At least, the test won't be conclusive-conclusive.
I could alternatively transform my data, and use a straight OLS regression, then report an R^2 for that approximation. Rich mentioned taking the square root of the data, which is the first time I've heard of using this transform for count data. Is that choice of transform a response to: O?Hara, R. B. and Kotze, D. J. (2010), Do not log-transform count data. Methods in Ecology and Evolution, 1: 118?122. ?