This paper presents patterns of adjacency in performance data and in cross-linguistic grammatical conventions. It is argued that a common principle of processing efficiency explains both: the more syntactic and semantic relations whose processing domains are minimized, and the greater the minimization preference in the processing of each relation, the more adjacency we find. The preferences of performance are quite systematic and it is suggested that they are ultimately motivated by reductions in simultaneous processing demands in working memory. The correlations with patterns of grammatical variation exist because grammars have conventionalized the adjacency preferences of performance.