This paper describes real user assessment of NativeAccentTM, a pronunciation tutor using automatic speech recognition that is a commercial product. It describes the product and discusses the issues involved in assessments of real users in real situations, such as assessments based on the customers own criteria instead of more academic measures, and the variations in the customers measures. Results in one study show that subjects who used NativeAccentTM did more than twice as well as the control group while both groups had human instruction. The implications of these results are discussed in light of other measures and real world considerations.