The ability to use the relations between visible landmarks to locate nonvisible goals (allocentric spatial coding) underlies success on a variety of everyday spatial orientation problems. Little is known about the development of true relational coding in infancy. Ninety-six 6-, 8.5- and 12-month-old infants were observed in a peekaboo paradigm in which they had to turn to a target location after displacement to a novel position and direction of facing. In a landmark condition, the target position was located between two landmarks, contrasted with a control condition in which no distinctive landmarks were provided. Six-month-old infants performed poorly in both conditions, 8.5-month-olds were significantly better with the landmarks, and 12-month-olds solved the task with or without landmarks. A follow-up study confirmed that the 8.5-month-olds used both landmarks to solve the task. This demonstration of allocentric spatial coding in 8.5-month-old infants shows earlier competence than that found in previous work in which only infants at the end of the first year were able to use landmarks relationally.