Three experiments examine whether a naturalistic reading task can induce long-lasting changes of syntactic patterns in memory. Judgment of grammatical acceptability is used as an indirect test of memory for sentences that are identical or only syntactically similar to those read earlier. In previous research (Luka & Barsalou, 2005) both sorts of sentences were found more acceptable, yielding a "structural preference" for recently encountered syntactic patterns. The effect is evident across a very diverse set of some 60 distinct sentence constructions. This phenomenon may be similar to implicit learning paradigms using evaluative ratings as dependent measures, but also resembles a variety of shorter-lived structural priming effects observed for a handful of syntactic constructions that occur in binary alternate forms. Here we examine the duration of the structural preference effect to determine whether it is fleeting (suggesting transient priming) or long-lasting (suggesting enduring changes in representation). Experiments 1 and 2 show that the effect is evident after a 48-h delay between initial exposure (simple reading aloud) and test. Experiment 3 shows a persistent effect after 7 days, despite unrestricted exposure to natural language in the interim. Larger effects were observed for comparatively novel sentence constructions than common syntactic patterns. Experiments 2 and 3 also examine the role of the evaluation process itself, and find that reading for comprehension produces greater facilitation than reading while performing an evaluative categorization (grammaticality rating). These observations suggest that incremental adjustments to the language processing system occur on a continuous basis and may extend to acquisition of novel syntactic structures. We discuss whether and in what ways the structural preference effect can be integrated with the literature on structural priming. We interpret our results in the context of models of language comprehension, emphasize the dynamic nature of grammatical knowledge in memory, and argue that the current paradigm offers several advantages for elaborating the various systems of memory upon which linguistic representations depend. (Contains 4 figures and 3 tables.)