The first Rasmussen report made the calculation that there would be one very large scale reactor accident in 100,000 "reactor years". That was without calculating human error. The second Rasmussen report including human error (but not bad human intention) came up with a roughly 1 accident in 50,000 "reactor years". Some German critics said the number may well be as low as 10,000 "reactor years".

We have one single datum so far for large scale accidents, Chernobyl was after roughly 4,000 "reactor years".

Let's ignore the critics and this single datum, for that reactor was a particularly bad design (I still hear the then West German communists saying that socialist reactors were inherently safer than capitalist reactors because they belonged to the people, but I digress).

Let us assume, 50,000 is a good guess. A "reactor year" is a year during which one reactor runs uninterrupted. So, if 100 reactors run one year without standstill, that makes 100 "reactor years".

In 2003, there have been 437 commercial reactors worldwide (noone knows exactly how many noncommercial reactors there are, so lets ignore them at this point). I therefore calculate roughly 400 "reactor years" per year. That makes 24,000 "reactor years" in 60 years (assuming the number of reactors is constant, which at the moment underestimates the number of reactors).

That is, even if we take the numbers from the second Rasmussen report which have been disputed and do not even try to calculate bad will actions, we have a very roughly 50:50 chance of seeing a very large scale accident within the next thirty years, somewhere in the world.
If the number of reactors would increase by factor ten (some energy scenarios say that) the number of "reactor years" per year will also increase by 10. With 5,000 reactors all over thw world, the pro-nuclear-energy Rasmussen report would expect us to adapt to roughly one very large scale accident some place in the world per decade.

These are in rough estimation the numbers as the nuclear industry itself deals with. The evaluation is of course up to yourself.
2015 (China), 2025 (Pakistan), 2035 (Iran), 2045 (Mexico), 2055 (?)

Wolfgang