The existence of a set of alternative models, all competing to characterise respiratory controller structure, leads naturally to a model discrimination problem. This paper considers the problem of input design for enhancing model discrimination. Both empirical and formal methods are considered. The empirical methods utilise frequency domain model characteristics. The formal methods utilise the time domain model responses. Using a class of models given by linear differential equations with time delays, an input consisting of a sum of sinusoidal components under a power constraint is considered. A procedure for designating the power spectrum of the signal is developed. Furthermore, a procedure for obtaining a binary signal with a similar spectrum is illustrated. These concepts are applied to input design for discrimination among models that characterise respiratory control during exercise.