Dynamic stability, as the degree to which identified segments at a given time remain unchanged over time in terms of number, size and profile, is a desirable segment property which has received limited attention so far. This study addresses the question to what degree latent classes identified from a stated preference choice experiment for environmental food claims remain stable over time. Following a cross-sectional design, two different consumer samples from an online panel provider participated in two identical choice experiments within eighteen months. On the aggregated level, a heteroskedastic logit model suggests significant changes in the price sensitivity and the utility from environmental claims between both experimental waves. A pooled scale adjusted latent class model is estimated jointly over both waves and the relative size of latent classes is compared across waves, resulting in significant differences in the size of two out of seven classes. These differences can largely be accounted for by the changes on the aggregated level. The relative size of latent classes is correlated at 0.52, suggesting a fair robustness. An ex-post characterisation of latent classes by behavioural and sociodemographic variables results in a large number of similarities of latent classes over time. Overall, latent classes seem to be relatively stable in their size and characterisation over time.