While articles are obligatory before singular nouns in many languages, in newspaper headlines they are frequently omitted. The optionality of article omission raises the question of why some articles are omitted and others not? We hypothesize that information-theoretic principles constrain the distribution of article omission, and specifically that it is more likely the more predictable the corresponding noun is. We tested this hypothesis with a rating and a corpus study on German newspaper headlines. Both confirm our hypothesis: Article omission turns out to be more likely (corpus study) and more strongly preferred (rating study) for relatively predictable nouns.