This talk weighs the trade-off between early cycle unit and component testing with late cycle integration and system testing and argues where the largest quality gains can be made by further investment in one, the other, or both. A very counter intuitive conclusion is reached and a model for testing that follows this conclusion is proposed.