This article aims at investigating the impact of financial supports from agricultural policy on farm-size dynamics. Since not all farms may behave alike, a non-stationary mixed-Markov chain modelling (M-MCM) approach is applied to capture unobserved heterogeneity in the movements of farms across economic size (ES) classes. A multinomial logit specification is used for transition probabilities and the parameters are estimated by the maximum likelihood method and the Expectation-Maximisation (EM) algorithm. An empirical application to an unbalanced panel from 2000 to 2018 shows that French farming consists of 'almost stayers', with a high probability of remaining in the same ES class over time, and 'likely movers', which present a higher probability of a change in size. The results also show that the impact of subsidies and other economic factors depends greatly on the type that a farm belongs to. These findings confirm that individual characteristics of farmers may be relevant for policy efficiency and more attention should thus be paid to unobserved farm heterogeneity in both policy design and the assessment of their impacts on farm-size dynamics.