The cognitive underpinnings of human recognition memory have been the subject of intense debate centered around two dominant accounts: the dual-process and the single-process model (Mandler, 1980; Yonelinas, 2002; Eichenbaum et al., 2007). The dual-process account posits that recognition implies two sequential operations: familiarity and recollection (Curran, 2000; Ranganath et al., 2004). Familiarity reflects stimulus knowledge that is assumed to be automatic, fast, threshold-based, and devoid of spatio-temporal details (Yonelinas, 1999). In contrast, recollection involves the conscious and deliberate retrieval of specific details or events from memory (Yonelinas, 2002; Brezis et al., 2017). Among the many tasks used to distinguish between the two processes (e.g., Dobbins et al., 2004; Ranganath et al., 2004; Yonelinas et al., 2019) the remember/know (R/K) paradigm is of particular interest as it allows participants to indicate whether they have access to contextual information (remember) or merely know that the item was presented (know; Düzel et al., 1997).

A major support for the dual-process account of recognition memory is provided by studies using event-related potential (ERP) techniques (Curran and Cleary, 2003; Vilberg et al., 2006; Woodruff et al., 2006). A repeated finding of these studies is that R/K judgments are associated with distinct ERP components over mid-frontal electrodes (“know”: FN400 component) and posterior electrodes (“remember”: Late Positive Component, LPC; for a review see Rugg and Curran, 2007). This temporal separation between R/K judgments, which are interpreted as indicators of recollection vs. familiarity, has been replicated numerous times with ERP (Curran and Cleary, 2003; Vilberg et al., 2006; Woodruff et al., 2006) or functional imaging studies (fMRI; Scalici and Caltagirone, 2017). Further investigations have shown that the two ERP components are differentially affected by manipulations of attention (Curran, 2004), confidence levels (Yonelinas, 2001), or memory performance in amnesia (Aly et al., 2010; Addante et al., 2012). R/K judgments identified with fMRI are associated with neural activations in two distinct regions of the frontal cortex. These findings indicate temporal, spatial, and functional differences associated with R/K judgments, which together support dissociated cognitive processes (Hill and Windmann, 2014; Hoppstädter et al., 2015; Andrew Leynes et al., 2019).

A less abundant line of research supports the idea that recognition memory relies on a single process. For example, some authors argued that the dissociation between R/K responses is confounded with the strength of the memory trace (Figure 1A; Finnigan et al., 2002; Brezis et al., 2017). According to this proposal, ERP components and time windows might be interpreted as two extremes on the continuum of a single variable: memory strength (Wixted and Stretch, 2004; Wixted, 2009). Some findings also suggest that familiarity and recollection are not stochastically independent regarding their contribution to recognition memory (Moran and Goshen-Gottstein, 2015). The single-process model has stirred considerable debate among scientists, and it is important to note that several subsequent studies have refuted this theory (e.g., Yu and Rugg, 2010; Addante et al., 2012; Addante, 2015). Some reports participating to the debate regarding the cognitive processes underlying recognition memory focused exclusively on correct recognition (i.e., hits and correct rejections; Wais et al., 2008; Hoppstädter et al., 2015), and thus provided arguments based on a biased view of recognition (Wixted, 2009). Several studies considered wrong responses (i.e., misses or false alarms; Curran, 2000; Rugg and Curran, 2007; Wolk et al., 2007; Addante et al., 2023), but a systematic analysis comparing all response types (i.e., correct and wrong responses) in a single analysis that englobes the entire electrode set (instead of focusing on selected electrodes) is missing. Within the framework of Signal Detection Theory (SDT), memory strength elicited by a stimulus represents a value on two overlapping distributions (“signal” and “noise”) that may be attributed to one of four possible outcomes: hit (seen and recognized), miss (seen but not recognized), correct rejection (CR; not seen and not recognized) and false alarm (FA; not seen but recognized). According to SDT, items are judged as “old” if the underlying signal strength exceeds an individual criterion (C; see Figure 1B). The interest in using such an approach is that outcomes can be decomposed according to behavioral response (old vs. new) or ground truth (signal vs. noise). Interestingly, R/K responses can also be considered as criteria on the continuum of memory strength (Donaldson, 1996; see Figure 1A). The R/K procedure is then compatible with an analysis of recognition memory within the SDT framework (Dunn, 2004).

Therefore, the question of interest when considering SDT to understand memory strength and decisional factors as core determinants of recognition memory is whether the four response outcomes generate specific ERP signatures. While some studies have examined memory strength as a moderating factor of ERP components (Brezis et al., 2017), a model-free investigation of ERP signatures of SDT outcomes is lacking. ERP correlates of incorrect responses (misses and false alarms) are less commonly studied, mainly because most previous studies used relatively simple recognition paradigms that did not generate a sufficient number of incorrect answers. For example, several studies used words, non-words, or simple images, which often produce recognition rates exceeding 80% or even 90% (Brady et al., 2008; Brezis et al., 2017; Delorme et al., 2018). These types of stimuli are not sufficiently complex to elicit a substantial number of wrong answers while still maintaining above-chance performance. Some authors have analyzed certain comparisons of responses (Leynes et al., 2005). However, what is missing from these studies is a robust analysis leading to the identification of the regions of interest in terms of temporal windows as well as in spatial/electrodes of maximal discrimination based on the response criterion. With this gap in the literature, the role of response biases and decisional factors in recognition memory remains uncertain. In this study, we applied an old/new task to test recognition memory of natural stimuli that were presented for a limited amount of time to increase the difficulty of the task. The study aimed to analyze old/new ERP data without a priori assumptions regarding time windows or regions of interest and without favoring a particular theoretical framework.

Our study focused on SDT parameters to investigate ERP correlates of recognition memory. By manipulating the complexity of the task, we obtained a sufficient number of wrong answers, i.e., FA and misses, to permit analysis of all four possible outcomes, while maintaining performance of all participants above chance. The waveform analysis identified one-time window of interest: 470–670 ms post-stimulus onset, and two clusters of electrodes: fronto-central and posterior-left. Comparisons of ERPs within each cluster showed that all SDT outcomes were well discriminated in the posterior-left cluster. In the fronto-central cluster, however, ERPs reflecting the same behavioral response, i.e., “yes” response (Hit and FA) vs. “no” response (CR and Miss) yielded indistinguishable ERPs even though they had different history: seen (Hit and Miss) vs. unseen (FA and CR). Additionally, we found a significant negative association between the discrimination index (d’) and ERP activation only in the posterior-left cluster. Our findings show that studying SDT outcomes expands our understanding of the electrophysiological correlates of recognition memory.

Behavioral performance in recognition memory often exceeds 80% or even 90% in typical old/new paradigms, indicating that human recognition memory is highly efficient (Manns et al., 2003; Brady et al., 2008; Brezis et al., 2017; Delorme et al., 2018). In our study, recognition performance was considerably lower (about 60%), which might be due to several factors. Many previous ERP studies used simple black and white images or (non-) words (Finnigan et al., 2002; Yu and Rugg, 2010), which have low ecological value and are highly discriminable (Felsen and Dan, 2005; Pinto et al., 2008) as compared to the photographs used in our study. Second, the short presentation time of 750 ms was likely insufficient for a thorough analysis and consequently precluded elaborated encoding of the stimuli. Superficial processing and low discriminability may favor item misses and false positive responses in our task, while previous studies of recognition memory generally only observed few item misses and even fewer FAs (e.g., MacKenzie and Donaldson, 2007; Hoppstädter et al., 2015).

Previous ERP studies testing recognition memory with the R/K paradigm distinguished two time windows that were, respectively, linked to familiarity (at 300–500 ms over fronto-central electrodes) and recollection (400–600 ms over left-parietal electrodes; Rugg and Curran, 2007). In contrast, our ERP analyzes only identified a single time window; that appeared to be temporally coherent with the late ‘parietal’ time window (470–670 ms) identified in previous studies (Curran, 2004; Rugg and Curran, 2007). The crucial question is whether the time window in our study reflects the same underlying cognitive processes related to recognition memory as proposed by other authors. Rugg and Curran (2007) summarized the findings of several studies by concluding that the late parietal component (LPC) is indicative of memory recollection. Their reasoning was based on the observation that the LPC was modulated by successful vs. unsuccessful source judgments or remember vs. know responses. Some also argued that the LPC was not related to response confidence or stimulus probability (Yu and Rugg, 2010). However, some of the findings supporting these conclusions are derived from studies that did not systematically compared correctly and incorrectly classified old and new items. This presents a problem for the interpretation of any electrophysiological component: while missed items may be explained by a failure of familiarity or recollection, it is difficult to explain FAs without recurring to alternative explanations. Examining only correct responses also ignores the fact that recognition memory reflects a decision-making process, particularly when subject’s confidence is weak.

Though the parietal time window reported in previous studies was identified by comparing seen to unseen stimuli, we observed a similar window after integrating all four SDT outcomes in an ANOVA. The slight temporal shift of approximately 70 ms might be explained by increased difficulty of our task, which may have delayed decisions due to uncertainty regarding items held in memory (Murata et al., 2005). We also identified a left-parietal cluster of electrodes that appears to be predominantly activated during this time window. One possible interpretation of this ERP components within our time window could be similar to the LPC described in previous studies, and therefore might reflect related electrophysiological processes. However, as no remember/know ratings were included in the current paradigm a direct comparison between our findings and research using the R/K paradigm is difficult and might be addressed in future studies. This information could be obtained by adding confidence scales to allow more granularity of the old-new paradigm, while still allowing the analysis of all SDT categories.

In contrast to previous work, we also observed a fronto-central electrode cluster that was active during the same time window. Our findings are not compatible with a single-cognitive process taking place in this time window, such as recollection, but rather suggest two distinct contributions to recognition memory. When examining ERP differences across conditions, we found that the posterior-left electrode cluster not only distinguished SDT outcomes according to item history (i.e., whether items had been presented before), but also the judgment of the subject (i.e., whether the subjects thought having seen the item before). In contrast, the fronto-central cluster only distinguished SDT outcomes according to the belief of the subject of having seen the item before. The distinction between item history and the subjective judgment of the observer is important, and can only be captured when all four SDT outcomes are considered.

The main strength of SDT is that it conceives item detection as a decision-making process, whereby subjective assessments, or beliefs, as well as decision criteria, come at play. Focusing only on ERP correlates of correct identifications of old vs. new items may identify electrophysiological processes underlying assumed memory components (such as recollection), but neglects the fact that memory is subject to metacognitive judgments. Our findings therefore open the debate as to which cognitive components are reflected by the identified ERP components. We would argue that in our findings associative strength of memory traces (Brezis et al., 2017) and metacognitive judgments partly dissociate across the posterior-left and the fronto-central electrode cluster. This hypothesis is based on the observation that the latter cluster only exhibited activity differences that could be explained by the type of response (yes vs. no), but not the actual item history. This finding is difficult to reconcile with the proposal that this cluster is specifically linked to familiarity (Hoppstädter et al., 2015), since FA items that were not seen before and should therefore not be familiar were processed similarly to target pictures. Also, interpreting this finding as reflecting the strength of memory traces requires the assumption that some items may have memory traces although they were never seen before. A more plausible possibility is that subjects decide whether they have seen an item before based on a comparison with the item pool, which becomes more difficult the more similar items have been presented.

Our finding thus suggests that the fronto-central cluster is specifically linked to metacognitive processes that guide the decision to produce a yes- or no-response. It is somewhat less straightforward to interpret the meaning of electrophysiological activity extracted from the posterior-left cluster, since this cluster differentiated between item history, but also between behavioral responses. Following our reasoning that responses in a recognition memory task reflect the strength of the memory trace and decisional processes, the posterior-left cluster appears to be a better predictor of performance. This conclusion is also supported by the correlation analysis, which showed that only ERP activity in the posterior-left cluster significantly predicted sensitivity in recognition memory (d’).

To sum up, our study shows that items in a recognition task do not only elicit different responses based on their representation in memory, but are also subject to complex decision processes. Such decision processes might operate on associative information, or on memory strength, which according to some authors may explain results of R/K paradigms within a single-process model (Brezis et al., 2017). However, ERP data extracted from the posterior-left cluster identified in our study can better be reconciled with a decisional process, rather than with the strength of memory representations. Memory strength predicts an arrangement of amplitudes that places hits and misses together (high memory strength), as opposed to CR and FA outcomes (low memory strength). Instead, Figure 5 shows that ERP amplitudes were arranged following the order Miss>CR > FA > Hit, which suggests that outcomes with yes-responses demand lower activations. This pattern suggests that the driving factor in the posterior cluster is not memory strength per se, but rather a metacognitive representation motivating a yes- or no-response.

In conclusion, by applying a signal detection framework we observed that the representation of items in memory (i.e., memory strength) as well as decisional processes affect the electrophysiological correlates of recognition memory. Based on our findings, we argue that classifying outcomes according to SDT enhances the possibility to analyze the electrophysiological components of recognition memory. Our study underlines the necessity to consider wrong answers (FA and misses) when analyzing recognition memory, as they may contain important information about the mental processes underlying the functions of human memory.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by Ethics Committee of the Canton of Geneva (approval number 2021-00414). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

SS: Conceptualization, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing. SC: Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Visualization, Writing – review & editing. AS: Funding acquisition, Resources, Supervision, Writing – review & editing. RP: Conceptualization, Resources, Supervision, Validation, Writing – review & editing.