# Accuracy of atomic clocks

When you hear about atomic clocks, it's accuracy is described by saying something like, " it neither gains or loses $x/y$th of a second in $z$ years." How is this error calculated? Does an error imply that we have a more regular physical phenomena with which to compare the atomic clock? Or does this error simply mean there are quantum phenomena that occur extremely rarely in the clock that might throw off its regularity every once and a while?

-