Models of ion channel kinetics subserve inferential methods applied to patch clamp data. For Markov models the density function of a sojourn time in a class of states is a mixture of exponentials. Determination of kinetic parameters from density functions may be complicated by non-uniqueness of solutions. This non-identifiability is investigated analytically for a class of two states, assuming detailed balance; relations between model properties, observable density parameters, and non-uniqueness are presented. The results are further developed in terms of similarity transform methods. Additional information provided by joint distributions is discussed. An example is given where identifiability of a model can be demonstrated explicitly. Attention is drawn to instances where the number of components in a density function may be misleading when used to infer the number of underlying states.