How many hidden neurons???

I've just builded a neural net with 2 input and 1 output neurons, the
hidden layer works fine with 20 neurons. The object is to predict
porosity values from gamma ray and electrical induction inputs. I've
been asked to explain why did I used 20 neurons in the hidden layer.
Is there some mathematical background or empirical relation to
determine the number of these neurons? Less neurons give poor results
and the more neurons I add the slower the program runs.
Thanks in advance

Manolo