Recent research on nearly 27 million scientific papers since 1950, and more than 5 million U.S. patents since 1970, shows that how scientists sample the ever-expanding literature is critical to making breakthroughs, irrespective of discipline. First, papers or patents that cite literature of a certain age range (mean of about 5 years, with a high variance) are roughly twice as likely to be a hit (in the top 5% of citations) than a field's average paper. By contrast, the 75% of papers and patents that combine knowledge from outside this “hotspot,” referencing the most popular or most recent literature—work commonly sampled by search engines —are not particularly likely to be hits. Second, hit papers mix highly typical and highly novel ideas and do so roughly according to a 90/10 ratio. Such papers referencing highly familiar knowledge—literature that historically has been cited together much more frequently than expected by chance—while at the same time citing papers that have rarely been co-cited before are at least twice as likely to be hits in their field than the average paper. Novelty is prized in science but becomes especially influential when paired with familiar, conventional thought. Sampling the literature also depends on team work. Team-authored papers are more likely to draw on work in the 5-year-old hotspot and to insert novel combinations into familiar knowledge domains than papers by solo authors. Such insights take us closer to uncovering approaches for searching for and recombining yesterday's ideas into tomorrow's acclaimed discoveries.