Many speech processing systems are commercially available and several applications have been successfully set up in different areas. In order to generalise their use in man-machine interaction applications a ``predictive model of performance '' of the technology to be used needs to be derived by application developers as a function of what they judge as relevant factors. This model of system performance, correlated to the technology specifications, has to be gauged against the ultimate target which is the application requirements and user needs. This chapter is based to a considerable extent on previous publications by [Choukri et al. (1988)] and Moore [Moore (1988), Moore (1994a)].