Well, what I said is a simplified version. The proper version is akin to Einstein's, I think:
"entities should not be multiplied beyond necessity."
And, to explain it a little:
"Furthermore, when multiple competing theories have equal predictive powers, the principle recommends selecting those that introduce the fewest assumptions and postulate the fewest hypothetical entities. It is in this sense that Occam's razor is usually understood."
Of course, like I said, it's not perfect. But when you have a system with an unknown number of variables, the solution which includes the least number of unknowns while still covering all cases is probably the best one.