Maximum Likelihood versus Goodness of Fit

If we have an i.i.d. sample, which can be modeled using two different families of distributions, how do we choose the best one:

http://www.r-bloggers.com/maximum-likelihood-versus-goodness-of-fit/