This article investigates the impact of agricultural policies on structural change in farming. Since not all farmers may behave alike, a non-stationary mixed-Markov chain modeling (M-MCM) approach is applied to capture unobserved heterogeneity in the transition process of farms. A multinomial logit specification is used for transition probabilities and the parameters are estimated by the maximum likelihood method and the Expectation-Maximization (EM) algorithm. An empirical application to an unbalanced panel dataset from 2000 to 2013 shows that French farming mainly consists of a mixture of two farm types characterized by specific transition processes. The main finding is that the impact of farm subsidies from both pillars of the Common Agricultural Policy (CAP) highly depends on the farm type. A comparison between the non-stationary M-MCM and a homogeneous non-stationary MCM shows that the latter model leads to either overestimation or underestimation of the impact of agricultural policy on change in farm size. This suggests that more attention should be paid to both observed and unobserved farm heterogeneity in assessing the impact of agricultural policy on structural change in farming.