KurisutinaJustin S. Feinstein, Melissa C. Duff, and Daniel Tranel. Sustained experience of emotion after loss of memory in patients with amnesia. PNAS 107(17), 7674–7679. Published online 12 April 2010; issue date 27 April 2010. Peer-reviewed primary human experiment. DOI verified: 10.1073/pnas.0914054107. Full primary text; PubMed metadata.
Evidence from this source
Five severely amnesic patients and five controls, individually matched on sex, age, and education, watched sadness-inducing films; four patient–control pairs later completed a happiness induction. Each induction used eight clips over approximately 17–19 minutes. Recall and verbal/picture recognition were tested 5–10 minutes afterward; four emotion assessments extended to approximately 20–30 minutes after the films. Patients reported elevated emotion after the memory test despite markedly impaired recollection. One patient recalled no sad-film details, failed to exceed chance on recognition, and reported sadness beyond 30 minutes. Others retained some recognition, sometimes substantial. The selected cohort had hippocampal amnesia; MRI volumetry was available for four patients, showing hippocampal reduction without significant amygdala-volume reduction. Sadness always preceded happiness on separate visits. Emotion trajectories were presented descriptively, without formal group tests because samples were small and variable. The finding concerns continuing reported affect over minutes; it does not establish durable learning, a retained event representation, or unconscious emotional experience. Paper, Methods, Table 1 and Figures 3–5.
Original interpretation and alternative accounts
For a functional model, remembering what happened and carrying a current state forward should be separate possibilities. A system that derives every present feeling solely by retrieving an explicit episode could miss a state that continues after that episode becomes difficult to access. A system that permanently stores every transient state as a stable disposition would make the opposite mistake. The paper motivates testing distinct time scales; it does not identify the biological storage medium or a uniquely correct architecture.
The dissociation is between reported present experience and performance on particular event-memory tests. Failure to exceed chance does not prove that every relevant representation is absent. Retained gist, implicit associations, bodily feedback, contextual cues, or other consequences of the induction could contribute. These alternatives can coexist with the central observation that rich explicit recollection is unnecessary for continued emotional reporting. Self-report here is the measure of conscious experience; calling the result “unconscious emotion” would reverse the distinction under study.
A material procedural alternative is that the memory test may itself renew affect. Recall and recognition probe the inducing events before the critical later emotion rating. Consequently, continuing state and test-triggered reinstatement are not cleanly separated. This is a possible confound inferred from the protocol, not a demonstrated explanation of the results. The detailed supplementary recognition procedure was unavailable, so the extent of such re-exposure cannot be quantified here. Memory-test difficulty or awareness of failure might also alter affect; positive as well as negative inductions make a simple frustration-only account less compelling, but do not replace a neutral-condition control.
The small clinical sample and fixed induction order cannot establish a general negative-versus-positive difference in decay or show that memory loss causes prolonged distress. Likewise, the findings do not demonstrate a durable cue–response disposition, recurrence after recovery, altered future choices, or persistence across days. Preserved affect is not evidence that the full experience, identity, or consciousness could be extracted and transferred.
Proposed functional-transfer experiment
Compare three predictive representations with matched observations: accessible event content alone, a short-lived affective-state model, and event content plus state. Include a simple autoregressive or state-space comparator, rather than attributing any gain to a language model. Use held-out future measurements and contexts to test whether observed recent affect improves prediction beyond what the person currently recalls. Keep present feelings, event/source memory, and later choices as distinct outcomes.
To distinguish persistence from renewal, counterbalance benign positive and neutral inductions, separate memory-probe and no-probe conditions, and vary when probes occur. Test delayed cue responses only after the immediate state has returned toward baseline. That creates a distinction between ongoing carryover and a disposition that can be reactivated later. Longer follow-up is required before calling either a durable memory. An engineered state variable would be an empirically assessed functional approximation, not a transferred feeling. No experiments or sessions were run for this note.
Reading and acquisition record
Read the complete main introduction, results, discussion, methods, Table 1, and figure captions from the retained public HTML. Checked the six-page PNAS early-edition PDF and visually inspected its main Table 1 and Figure 4. Graphical values were not independently digitized or reanalyzed. No raw data, code, or individual participant records were accessed; bibliography entries were not followed individually.
Supplementary SI Text, Tables S1–S4, and Figure S1 were identified but not retrieved or read. Thus, detailed lesion histories, recognition-test construction, composite calculations, film lists, and individual happiness trajectories remain incompletely checked. PMC's linked supplement returned a download/challenge HTML page; publisher and Europe PMC mirror attempts failed; the Europe PMC supplementary API explicitly declined this non-open-access item. Failed responses were not retained as source documents. The main paper is publicly readable, but no open reuse license was identified.
Local sources: full HTML, readable text, early-edition PDF, provenance and SHA-256 hashes. Accessed 22 September 2026. The PDF was retrieved from a public educational copy hosted by Scripps Osteopathic Medical Center; primary-text claims were checked against PMC.