The car sequencing problem involves scheduling cars along an assembly line while satisfying capacity constraints. In this paper, we describe an Ant Colony Optimization (ACO) algorithm for solving this problem, and we introduce two different pheromone structures for this algorithm: the first pheromone structure aims at learning for ``good'' sub-sequences of cars, whereas the second pheromone structure aims at learning for ``critical'' cars. We experimentally compare these two pheromone structures, that have complementary performances, and show that their combination allows ants to solve very quickly most instances.