Psychological Bulletin 1986. Vol. 99. No 3. 3:0-329 Copjnght 1986 bs [he American Psychological Association. Inc. 0033-:909/B6/S00.75 Eyewitness Identification: The Importance of Lineup Models Gary L. Wells and John W. Turtle University of Alberta. Edmonton. Alberta, Canada A detailed analysis is made of lineup models for eyewitness identification. Previous treatments of eyewitness identification have nol distinguished between the all-suspecl model and the single-suspect model. The single-suspect model allows for the occurrence of foil identifications, a know n-error category, whereas the all-suspect model does not. A Bayesian analysis of posterior probabilities of the guilt of a given suspect under various prior probabilities shows that the all-suspect model maj be more or less diagnostic than the single-suspect model depending on the extent to which the use of suspects rather than foils increases the prior likelihood that the actual target is in I he lineup. On the other hand, the lineup-wise error rate (which is the likelihood that any suspect will be falsely identified) is considerably higher with the all-suspecl lineup. Field data show that the all-suspect lineup is sometimes used by police departments, and some data suggest that police do not appreciate the distinction between the two models with regard to lineup-wise error rates. It is recommended that either a mixed model or a preceding blank lineup be used to replace all-suspecl models in actual cases. Although eyewitness testimony received some attention from experimental psychologists over 75 years ago (Munsterberg, 1908), it has only been within the last decade thai significant efforts have been made to build a useful literature. Six recent books have been devoted entirely to the psychology of eyewitness testimony (Clifford & Bull, 1978; Lloyd-Bostock & Clifford, 1982; Loftus, 1979; Shepherd, Ellis, & Davies, 1982; Wells, & Loftus. 1984; Yarmey, 1979), and articles have appeared in almost every major psychological and legal journal. There now are hundreds of eyewitness experiments, many dealing with the issue of eye- witness identification. Despite experimental evidence showing relatively high rates of false identification, controversy exists over the extent to which false identification rates are also high in real-world cases, and, if they are high, what can be done about them (see Egeth & McCloskey, 1984; Loftus, 1983; Wells. 1984a). The experimental research shows false identification rates of 20% or even 60°<> under some conditions. Clearly, the rates obtained in these experiments are well above the false identification rates that are obtained in real-world cases. Yet this apparent discrepancy between exper- imental data and real-world cases is not likely to be attributable to differences in witnessing and testing conditions. For instance, most experiments use short delays between witnessing and testing (e.g., 5-20 min vs. days or even months in real-world cases). Also, experiments typically use college students whose mean age and general health are likely to make them better witnesses than the general witnessing population (Yarmey, 1984). Although many cases of rape or kidnapping can involve qualitatively and quantitatively superior encoding conditions than most experi- ments provide, experiments generally do a good job of mimicking This research was made possible by grants to the first author from the Social Sciences and Humanities Research Council of Canada and the University of Alberta. Correspondence concerning this article should be addressed to Gary L. Wells, Department of Psychology, P-220 Biological Sciences Building, University of Alberta. Edmonton, Alberta, Canada, T6G 2E9. the encoding conditions of common crimes such as robbery or theft. Finally, low levels of stress and good lighting characterize experiments, whereas actual cases frequently involve significant stressors and poor viewing conditions. Indeed, as experiments come closer to simulating the experience of actual eyew itnessing, identification accuracy seems to decline (Murray & Wells, 1982). No doubt there are some experiments that provide witnessing and testing conditions that are poorer than real-world conditions, but the apparent discrepancy between laboratory findings and real cases may be better explained at another level. We argue that the rates of false identification obtained in ex- periments might in fact be a fair representation of the frailties of eyewitness identification. This is not to say that false identi- fication rates in real cases are as high as 60°i or even 20%. As discussed in the next section, not all misidentifications qualify as false identifications, and there are base rates (or prior prob- abilities) that must be considered as well. The analyses that we present show how the likelihood of false identification can be quite low or quite high depending on the "lineup model" used in the case. (For the precise meaning of lineup model see the section on Single-Suspect Versus All-Suspect Models.) Lineup models have not been considered previously in the psychological or legal literature on eyewitnessing, yet they might account not only for the apparent discrepancies between laboratory eyewitness research and real-world cases, but also for differences between police jurisdictions in their rates of false identification. Before discussing models of lineups, two important distinc- tions must be introduced. These distinctions are (a) known versus unknown errors and (b) target-present versus target-absent lineups. Necessary Distinctions Known and Unknown Errors Most eyewitness researchers recognize immediately the fact that not all errors committed by eyewitnesses are equal in con- sequence. A false identification of a suspect from a lineup, for 320 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.