Is it possible to learn the relation between 2 nonadjacent events? M. Peña, L. L. Bonatti, M. Nespor, and J. Mehler (2002) claimed this to be possible, but only in conditions suggesting the involvement of algebraic-like computations. The present article reports simulation studies and experimental data showing that the observations on which Peña et al. grounded their reasoning were flawed by deep methodological inadequacies. When the invalid data are set aside, the available evidence fits exactly with the predictions of a theory relying on ubiquitous associative mechanisms. Because nonadjacent dependencies are frequent in natural language, this reappraisal has far-reaching implications for the current debate on the need for rule-based computations in human adaptation to complex structures.The idea that most higher cognitive activities, especially language comprehension and production, are based on abstract, rulebased operations on symbolic contents is one of the cornerstones of the mainstream tradition in cognitive psychology. However, over the last 2 decades or so, an alternative view has gained increasing influence. This alternative conception is rooted in the traditional associative view of mind, but its recent upsurge is essentially related to the growth of connectionist modeling (e.g., McClelland & Rumelhart, 1986). Indeed, connectionist studies have shown that certain activities that were once thought of as straightforward evidence for rule-based computations can be simulated by models that rely only on associative mechanisms. This issue has crystallized around two broad conceptions of the mind, often thought of today as an opposition between those who advocate the need for assuming algebraic-like computations (e.g., Marcus, Vijayan, Rao, & Vishton, 1999;Pinker, 1997) and the proponents of statistical/distributional approaches (e.g., Seidenberg & MacDonald, 1999). The English past tense has been a focus for this debate from its outset, and the number of papers pertaining to this issue over the last few years (e.g., Pinker & Ullman, 2002;Ramscar, 2002) suggests that it is still unsettled despite a considerable amount of research effort.A newcomer in this lively debate is the learning of nonadjacent (or remote) dependencies. The major part of the traditional literature on associative learning has dealt with relations between adjacent events. This is true both for the domain of animal conditioning and for studies on paired-associate learning in humans. In both cases, the items to be associated are displayed in close temporal or spatial proximity. The same is true for the more recent studies on implicit learning (e.g. Stadler & Frensch, 1998). Looking at a standard flowchart of a finite-state grammar commonly used in artificial grammar studies is sufficient to show that relations are built between contiguous elements. 1 Several studies have shown that those adjacent relations were far more relevant for linguistic structure than researchers had claimed in the past. For instance, Redington, Chater, and Finch (199...