Dinucleotide repeats in the Drosophila and human genomes have complex, length-dependent mutation processes.

We use methods of maximum likelihood estimation to fit several microsatellite mutation models to the observed length distribution of dinucletoide repeats in the Drosophila and human genomes. All simple models are rejected by this procedure. Two new models, one with quadratic and another with piecewise linear slippage rates, have the best fits and agree with recent experimental studies by predicting that long microsatellites have a bias toward contractions.