Chi square only tests the distribution of numbers in categoies. It isn't the only test to ensure some reasonable randomness. there's still the problem that the routine could genrate fifty 1's, fifty 2's, then fifty 3's etc. so that all the categories have the same number of hits, but the run sequence is nowhere near random. You also need some non-parametric runs test to look for combinations in a series to ensure the routine isn't biased. A chi square can show excellent results when the PRNG returns even, odd, even, odd, even, odd, etc.
Jack Peacock