Scientists long blamed infantile amnesia on an immature hippocampus that couldn't properly make memories. Then researchers scanned the brains of 26 babies and found the memories being encoded. The problem may not be making our earliest memories at all — it may be finding them again years later.

The result concerns encoding, not adult recall

A 2025 paper in Science found that activity in the infant hippocampus predicted which pictures babies would appear to recognize about a minute later. Tristan Yates and colleagues collected usable functional MRI scans from 26 awake infants aged 4.2 to 24.9 months.

This is one study, not settled consensus. It provides evidence that parts of the memory-making machinery are working during infancy, especially from around the first birthday. It did not follow a memory from infancy into adulthood or demonstrate that forgotten childhood experiences remain stored for decades.

That distinction narrows the old explanation for infantile amnesia. A young hippocampus may be capable of encoding at least some individual experiences well before people can later describe an autobiographical memory from the same period.

The experiment turned looking into a memory test

Testing episodic memory in a preverbal infant requires an indirect measure. The team showed each baby a new photograph of a face, object or scene for two seconds inside the scanner. After 20 to 100 seconds, with three to 12 other items intervening, the old picture reappeared beside a new one from the same category.

If a baby looked longer at the old picture, the researchers treated that as a familiarity preference. They then sorted the original viewing trials according to what happened at test. The question was whether brain activity during the first encounter differed for items that later received the longer look.

It did. Blood oxygen level-dependent activity, the indirect signal measured by fMRI, was higher in the hippocampus during encoding of pictures followed by a familiarity preference. The effect was clearest in the posterior hippocampus, a region involved in episodic encoding in adults, and among the 13 infants older than 12 months.

The babies did not, as a group, look at old pictures more often than chance.

The finding came from the relationship within individual trials between earlier hippocampal activity and later looking, not from an overall behavioral memory advantage across the sample.

A one-minute memory is not an autobiography

Encoding is the first stage in the life of a memory. Consolidation stabilizes it, storage preserves it over time, and retrieval makes it accessible later. Evidence for one stage does not establish that all the others succeeded.

The task also used isolated images. It did not ask whether infants remembered where an image appeared, what happened around it or how it related to another event. The authors describe the capacity they measured as possibly rudimentary. Rich autobiographical memories depend on contextual and relational information that this experiment did not test.

The fixed two-second exposure may also have affected younger infants differently from older ones, and 26 participants is a modest sample for tracing development across a 20-month age range. Functional MRI records changes related to blood oxygenation, not a memory trace itself.

Why retrieval remains a hypothesis

The finding weakens a broad claim that hippocampal immaturity prevents memory encoding throughout infancy. It does not identify what later makes early memories inaccessible. Rapid forgetting, incomplete consolidation, changing retrieval cues and developmental reorganization could all contribute.

Earlier work from the Yale group found hippocampal involvement in statistical learning in much younger infants. Statistical learning extracts patterns across repeated events. The new work adds evidence for one-shot encoding of a particular visual item, with the stronger signal appearing later in infancy.

A different line of evidence comes from animals. In a 2016 Nature Neuroscience study, memory traces formed in infant mice could be reactivated after the learned behavior was no longer expressed. That supports a retrieval account in rodents. It cannot establish that inaccessible autobiographical memories persist in adult humans.

An accompanying Science commentary called the newly demonstrated human memories fleeting. In my reading, that description keeps the central result in proportion: the study captured an early neural signature of short-delay recognition, not a sealed archive of babyhood.

What would test the missing years

The researchers now need to track known experiences over much longer periods. According to Yale’s account of the ongoing work, the team is testing whether infants, toddlers and children recognize home videos filmed from their own earlier perspective. Tentative pilot observations suggest accessibility may persist into preschool before fading, but those observations are not the result reported in the Science paper.

Retrieval cues can alter what adults manage to reconstruct, as ScienceBlog’s earlier look at music and autobiographical memory explains. A longitudinal infant study could use controlled cues to ask whether an apparently lost event is absent, weakly stored or simply difficult to reach.

For now, the experiment establishes a starting point.

Around one year of age, the hippocampus can participate in encoding an individual visual experience. Whether that trace survives the developmental changes that follow remains unanswered.