We introduce a new measurement of novelty based on the frequencies of pairwise combinations of article keywords. On the set of all research articles published from 1999 to 2013 in the journals... Show moreWe introduce a new measurement of novelty based on the frequencies of pairwise combinations of article keywords. On the set of all research articles published from 1999 to 2013 in the journals referenced by the WoS (more than ten million papers), we find no evidence of shrinking novelty in science over that period. Novel contributions are more often performed in larger teams that span more institutional boundaries and geographic areas. High novelty increases both citations and the odds of a “big hit” by more than forty percent but individual level returns to novelty may be low. In such a rapidly evolving environment, novelty is a lever for decreasing risk (not the reverse) as it increases the probability of addressing a problem that remains active in the future. We document that top journals play a very significant role in sustaining novelty in science. Show less