What is a “free parameter” in a computational model?

In many articles regarding computational models of some particular phenomenon, there seems to be a consensus: "the smaller the number of 'free parameters' in the model, the better". So, what is meant by "free parameter", and why is it less desirable to model something with such a parameter?

Thanks!

-
I don't know if this is really an appropriate question for MO (see the FAQ), but my understanding is that a free parameter is one not specified by the model which must be extrapolated from data. The more free parameters in a model, the greater the danger of overfitting to the data, and the less explanatory power the model has (think epicycles vs. ellipses). – Qiaochu Yuan Jul 3 2011 at 5:05