Factor oracles, built from an input text, are automata
similar to suffix automata, and
accepting at least all substrings of the input text.
In 2000 Lefebvre et. al presented an algorithm
to detect repeats on text with using factor oracles.
They presented improved version of this method in 2002.
Although repeats found with these methods are not maximal,
average error is very low and algorithm runs quite fast.

In this paper, we present two ideas to improve accuracy of these algorithms.
One is slightly modified version of algorithm presented in 2000,
but improves accuracy significantly, with slight overhead of running time.

The other uses new property of factor oracles shown in
the technical report C-185 (Dec 2003),
to detect maximal repeats of an input text (but slow).