Résumé : For more than two decades, scholars have been debating the so-called personalization of politics. Some studies confirm such an evolution, while others demonstrate that evidence of personalization is at best mixed, or even absent. This article aims at shedding a new light on this controversy by looking at the evolution of the use of preferential voting in Belgium. Preferential voting has been constantly growing, but since 2007, the trend has been reversed and fewer voters decide to cast a preferential vote. We argue that this decline is not evidence against personalization. Rather, it illustrates the need to distinguish conceptually and empirically between two dimensions of personalization: ‘centralized’ and ‘decentralized’ personalization. The decline in the use of preference votes is not related to a decline in the former (which refers to a handful of political leaders). Instead, it is due to the decline of the latter form of personalization (referring to a large group of individual politicians). Candidates other than party leaders appear to have growing difficulties to attract votes. This negative relationship holds, even when we control for measures of electoral reform and the newness of parties. Our results also show that leadership effects are stronger in new parties.