How do people make such rich inferences from such sparse data? Recent research has explored this inferential ability by investigating probabilistic reasoning in infancy. For example, 8- and 11-month-old infants can make inferences from samples to populations and vice versa (Denison & Xu, 2010a; Xu & Garcia, 2008). The current experiment investigates the developmental origins of this probabilistic inference mechanism with 4- and 6-month-old infants. Infants were shown 2 large boxes, 1 containing a ratio of 4 pink to 1 yellow balls, the other containing the opposite ratio. The experimenter sampled from, e.g., the mostly pink box, and removed either 4 pink and 1 yellow or 4 yellow and 1 pink ball on alternating trials. 6- but not 4-month-olds looked longer at the 4 yellow and 1 pink ball sample (the improbable outcome) than the 1 yellow and 4 pink ball sample (the probable outcome).