Determining causal effects

I have been wondering for a long time how one can determine causal relationships between A and B, instead of just correlations. Apparently, the method employed is that of “internal validity” (http://en.wikipedia.org/wiki/Internal_validity).

One take away message: If we intentionally manipulate A (independent variable) and observe the effects on B, that has stronger internal validity than if we just observe the values of A and B independently.