Bayesian models of starting price effects in contingent valuation studies have been used to track and to correct for starting price effects for more than two decades. Since then, the single parameter form that has been used to estimate the starting price effect has been adapted to distinguish among strategically induced, Bayesian or yea-saying responses (Whitehead, 2002). Recent works, however, show that these diagnostics can be inaccurate if heterogeneity exists in the value of the updating parameter. This work allows a more flexible functional form, beyond the single updating parameter, to capture a potentially richer set of starting price effects that may operate in a single dataset. First we employ a finite mixture model to sort respondents into latent classes. The finite mixture model locates three distinct types (latent classes) of respondents, and each exhibits a discretely different starting price-affected behaviour in a subsequent regression. Heterogeneity does not follow a continuous distribution across one starting price effect, such as a Bayesian updating parameter, that might be captured by a random parameter estimator. Rather effects fall into discrete classes. Researchers will need to better match theory to the diverse array of starting price effects detected.