Are random vectors useful in test generation?

Most automatic test generation systems start by generating random vectors, then switch to a fault-oriented algorithm which targets the remaining undetected faults. Experimental results are presented which show that, for most circuits, using random vectors increases both the total test generation time and the total number of tests. This is true not only for random-pattern resistant circuits, but also for circuits where random vectors easily achieve high fault coverage. Running only a fault-oriented algorithm that selects its next target as close as possible to primary inputs, and coupling it with random fill, is more efficient than a two-phase strategy using random vectors in phase one