Four random test generation strategies are compared to determine their relative effectiveness: equiprobable 0s and 1s; two weighted random pattern generation algorithms; and the maximum output information entropy principle. The test generation strategies are compared at a variety of target fault coverages. Two statistically based metrics are used to evaluate the techniques: a large-sample test of the difference of means and an upper confidence limit. The two weighted random test pattern generation strategies are found to be generally superior to equiprobable 0s and 1s and maximum output entropy. For a given logic circuit, the same technique is not necessarily optimal at every fault coverage.