This paper compares various ways of accounting for attribute non-attendance in a model of patients' choice of general practitioner appointment. The methods considered include the endogenous attribute attendance model, which is a variant of the equality-constrained latent class model, and various forms of mixed logit models with parametric and non-parametric coefficient distributions. It is found the the willingness to pay estimates derived from the models allowing for attribute non-attendance are considerably lower than those derived from a standard logit model.