Central to communication are acts of reference, in which speakers clarify to hearers the identity of specific individuals (referents) through the use of referring expressions. Consequently, pragmatic competence in referring is likely to be an important skill for second language learners (SLLs). One of the complexities that speakers face is that the most appropriate referring expression (RE) on one occasion may be entirely inappropriate for that same individual on a different occasion. To explore pragmatic appropriateness in the use of REs, two approaches appear particularly apt. Firstly, Accessibility Theory (Ariel, 1990) holds that speakers select an RE type based on their best estimate of how recoverable the referent is in the addressee’s memory. Secondly, acts of reference may be most helpfully viewed as occurring across discoursal units above the level of the noun phrase. However, it appears that neither of these frameworks has been adopted in previous studies of SLL reference. In addition, it appears that little has been reported in relation to the factors that trigger referential miscommunication. This study examined references by relatively advanced SLLs and by L1 participants. The focus was on pragmatic appropriateness in marking accessibility, and factors implicated in referential miscommunication. A film retelling task was used to elicit data from participants in twenty SLL-L1 and ten L1-L1 dyads; following each retelling, a two-part stimulated recall interview was conducted with the addressee to identify miscommunications. Individual acts of reference were analyzed according to an accessibility coding system. Miscommunications were analyzed and likely triggers identified in terms of, for example, accessibility marking and NP errors. Findings are reported in relation to SLL referential competence, triggers of referential miscommunication, and some more general issues relating to the nature of reference. When introducing hearer-known referents into discourse, the SLL participants tended to be under-explicit for entities with low-accessibility but over-explicit for central characters. The least accessible entities were frequently either avoided or were under-explicit and, for this reason, were often miscommunicated, suggesting that these learners had difficulties constructing references across larger discourse units. In subsequent references (referent tracking), the SLLs tended to be over-explicit, particularly in certain contexts. It appears that this resulted partly from a communicative strategy and partly from cognitive load. The most frequent trigger of miscommunication in referent tracking was pronoun errors. More generally, I argue that the present approach to miscommunication analysis complements error analysis. The theoretical framework for the study involves a core definition of reference (adopted from Bach, 2008), which is then relaxed in two stages, thereby distinguishing three levels of reference. I argue that this framework enables principled definitions of reference and referential miscommunication, which partially bridge contrasting views found in philosophy and linguistics. The findings challenge three widespread assumptions about reference: (a) that all references require resolution, (b) that interactants attempt to clarify all references, and (c) that only highly accessible referents are felicitously referred to with pronouns. I propose a theory of referentiality in which speakers are held to signal the extent to which references require resolution.