Here we present an expectation-maximization algorithm built on a pair hidden Markov model that is able to properly handle indels in neutrally evolving DNA sequences.