Retroactive Interference

Abstract

Retroactive interference is the disruption of an established memory by learning that occurs afterward: material acquired later reaches back to impair the recall of material acquired earlier. First isolated in the paired-associate laboratory, it became the central evidence that forgetting reflects competition among memories rather than the passive fading of a disused trace. The classic analysis decomposes it into two factors — competition among responses at retrieval and genuine unlearning of the original associations — and modern work adds a third: newly encoded material can interfere by disrupting the consolidation of what came before. This article traces the phenomenon from the interpolated-list experiments of the 1930s and 1940s to contemporary accounts framed around retrieval-induced forgetting, adaptive suppression, and systems consolidation, with interactive demonstrations of each.

Keywords: retroactive interference, unlearning, consolidation, forgetting, interpolated learning

Retroactive interference (RI) denotes the reduction in retention of previously learned material caused by activity — most often new learning — interpolated between original acquisition and the test of retention (McGeoch, 1932). It is the mirror image of proactive interference, in which earlier learning impairs the retrieval of what follows; together the two directions established interference, rather than decay, as the dominant mechanism of forgetting in verbal learning (Underwood, 1957). The phenomenon matters because it reframed the basic question of memory: why a trace is lost is, in large part, a question about what else was learned, not about how much time has passed.

What Retroactive Interference Is

The canonical demonstration uses paired-associate lists. An observer first learns a list of stimulus-response pairs — the A-B list, pairing each cue A with a response B. A second list is then learned in which the same cues are paired with new responses — the A-C list. When retention of the original A-B pairs is finally tested, recall is markedly worse than for a control group that spent the interpolated interval on an unrelated activity (Melton & Irwin, 1940). The interpolated learning has reached back and damaged the original memory; that backward-acting loss is retroactive interference.

The design isolates the interfering event. Two groups learn the same original list and take the same final test after the same delay; they differ only in what fills the retention interval. Any difference in final recall is therefore attributable to the interpolated learning itself, not to the passage of time. This logic — holding the interval constant and varying only its content — is what let the interference tradition separate forgetting-by-interference from forgetting-by-decay, a distinction that pure elapsed-time studies cannot draw.

Figure 1

The Interpolated-Learning Design for Retroactive Interference

Interference Versus Decay

Before the interference tradition, forgetting was widely attributed to disuse — a trace was thought to weaken automatically as time passed without rehearsal. McGeoch's decisive critique was that time itself is not a cause: nothing happens because of time, only during it (McGeoch, 1932). Rust corrodes iron over time, but the cause is oxidation, not duration; by the same logic, memories are lost over an interval because of what occupies the interval, not because the interval elapsed. The demonstration that filling a delay with similar learning produces far more forgetting than filling it with rest, or with sleep, made the point empirically. Jenkins and Dallenbach had shown as early as 1924 that observers who slept through the retention interval forgot far less than those who stayed awake for an equal period, precisely because sleep withholds the interpolated activity that waking supplies (Jenkins & Dallenbach, 1924). What decays is not the trace but access to it, and what governs that access is competition from other memories.

This reframing did not abolish the possibility of decay so much as demote it. Later work established that both directions of interference operate, and that proactive interference — the buildup of intrusions from earlier lists — accounts for a large share of the forgetting once attributed to time (Underwood, 1957). Retroactive interference remained the cleaner experimental handle, because the interfering event can be scheduled, measured, and withheld from a control group.

The Two-Factor Account

Melton and Irwin proposed that retroactive interference is not one process but two (Melton & Irwin, 1940). The first is response competition: at test, the newly learned C responses compete with the original B responses for the same cue, and intrusions of C where B was wanted are the overt sign of it. The second they inferred rather than observed. When they plotted overt C-intrusions against the amount of interpolated learning, competition rose and then fell, yet total interference kept climbing. Something beyond competition was suppressing the original responses; they named the residual Factor X, and it was later identified as genuine unlearning of the first-list associations.

Barnes and Underwood supplied the direct evidence with a modified-modified free recall (MMFR) test: after learning both lists, observers were asked to produce both responses to each cue, with unlimited time and no competition to resolve (Barnes & Underwood, 1959). If interference were only competition, removing the competition should restore the B responses. It did not — first-list responses were genuinely lost, and lost more the more C-learning had intervened. The unlearning was real, not a retrieval artifact. Briggs had already shown that the lost first-list associations spontaneously recover some strength over time, exactly as an extinguished conditioned response recovers, which anchored the analogy between unlearning and experimental extinction (Briggs, 1954).

Table 1. Three loci at which interpolated learning is held to impair an earlier memory. They are not mutually exclusive; a single episode of interference may act at more than one.

Similarity and the Amount of Interpolated Learning

Two variables govern how much a memory suffers. The first is similarity: interference grows as the interpolated material more closely resembles the original. McGeoch and McDonald varied the relation between the two lists and found the greatest retroactive inhibition when the interpolated responses were synonyms of the originals, less for unrelated words, and least for a resting interval (McGeoch & McDonald, 1931). Shared cues with different responses — the A-B, A-C structure — are the maximally interfering case precisely because the same retrieval route now leads to two destinations. The same principle explains interference in recognition: memory for an event is distorted by later, related events that overwrite overlapping details (Chandler, 1991).

The second variable is degree of interpolated learning. The more the second list is practiced, the more the first is lost — the relationship Melton and Irwin charted and the one the two-factor demonstration above reproduces. The two variables interact: a small amount of highly similar interpolated learning can do more damage than a large amount of unrelated material, which is why everyday interference is dominated by activities that reuse the same representations rather than by sheer mental busyness.

Consolidation and the Neuroscience of Interference

A newer locus of interference sits before retrieval altogether. A freshly encoded memory is labile for a period after learning, and consolidation is the process that stabilizes it; new learning that arrives during this window can disrupt consolidation and so cause forgetting that has nothing to do with response competition (Wixted, 2004). This idea is not new — Müller and Pilzecker proposed it in 1900 under the name perseveration — but it was revived when studies of amnesia and of ordinary forgetting converged on the same conclusion. Dewar and colleagues showed that reducing interference in the minutes after learning, by having people rest quietly rather than take in new material, sharply improves later retention, tying everyday forgetting to the same mechanism that spares memories in some amnesic patients (Dewar, Cowan, & Della Sala, 2007).

The protective effect of an unfilled interval is robust: a brief period of wakeful rest after learning preserves the fine detail of new memories that an equivalent period of new activity erodes (Craig & Dewar, 2018). The dependence runs both ways in time — the fate of a new memory depends on what is learned just before and just after it — and the interference is often specific to overlapping content and overlapping neural representations rather than general mental load (Robertson, 2012). On this view retroactive interference is not a single laboratory curiosity but a family of effects unified by competition for shared representations, some resolved at retrieval and some settled much earlier, while the trace is still being written.

Worked Example

Consider the two-factor demonstration with ten trials of interpolated A-C learning. The illustrative model sets baseline recall of the original A-B list at 85%, with two subtractive losses. Unlearning grows toward an asymptote of 50 percentage points as the second list is practiced; after ten trials it has reached 0.50 × (1 − e^−10/16) ≈ 23 points. Response competition follows a rise-then-fall curve that peaks near six trials and has begun to recede by ten, contributing 0.15 × (10/6) × e^(1−10/6) ≈ 13 points. Final recall is therefore 85 − 23 − 13 ≈ 49%, a total retroactive loss of 36 points from baseline.

The decomposition is the instructive part. At five trials the two factors are almost equal (competition ≈ 15 points, unlearning ≈ 13). By forty trials competition has nearly vanished — the second list is now so well learned that it rarely intrudes overtly — yet total interference is at its worst, because unlearning has climbed to about 46 points. This is exactly the pattern that forced Melton and Irwin to posit a second factor: overt competition cannot explain interference that keeps growing after the competition itself has subsided. The values are a schematic illustration of that logic, not measured data; real slopes vary with materials, spacing, and observers.

Discussion

Retroactive interference reoriented memory research from a question about time to a question about content, and the reorientation has held. The two-factor account remains the textbook backbone, but its second factor has been re-described more than once. Postman and Underwood folded unlearning into a broader theory in which a set of first-list responses is suppressed as a whole and must be recovered, and they catalogued the anomalies the simple account could not handle (Postman & Underwood, 1973). Anderson later recast the storage-side loss as an active, inhibitory process under executive control rather than a passive erosion, connecting classic interference to retrieval-induced forgetting, in which retrieving some items suppresses their competitors (Anderson, 2003).

The recovery data complicate any account that treats the first list as simply destroyed. Lustig and colleagues argued that much apparent unlearning is a bias in what is accessible at test — an over-reliance on the most recently strengthened route — rather than the outright loss of the original association, since controlled retrieval can bring first-list responses back (Lustig, Konkel, & Jacoby, 2004). Whether the original memory is erased, suppressed, or merely out-competed remains the live question, and it is not merely terminological: the three possibilities make different predictions about when, and whether, an interfered memory can be restored.

Current Directions

Neuroimaging has begun to watch interference happen. Retrieving a target memory measurably suppresses the cortical representation of its competitors, and the degree of suppression predicts how much those competitors are later forgotten — a neural signature of adaptive, retrieval-induced forgetting rather than of passive decay (Wimber et al., 2015). This has motivated a broader reframing of forgetting as adaptation: prefrontal control actively down-regulates memories that interfere, and forgetting is in part a feature the system uses to keep the useful memories reachable (Anderson & Hulbert, 2021).

A second front concerns the nonmonotonic effect of retrieval on neighboring memories. Moderate activation of a competitor can weaken it while strong activation strengthens it, a nonmonotonic plasticity rule that predicts when retrieval will hurt versus help related memories (Ritvo, Turk-Browne, & Norman, 2019). A third links interference to consolidation dynamics: retrieval can act as a fast route to stabilizing a memory, so that testing both protects the retrieved item and reshapes its competitors (Antony et al., 2017). Underlying much of this is evidence that different memory representations forget along different trajectories, so that whether interference bites depends on the format in which a memory is held (Sadeh et al., 2016). The through-line is that retroactive interference is being redescribed as one visible consequence of the mechanisms that keep memory adaptive.

Commonly Confused With

Proactive Interference

The prefix names what does the interfering, not what gets harmed. Retroactive: new learning reaches back to disrupt an old memory. Proactive: old learning reaches forward to disrupt a new one. Ask which material is the culprit — in retroactive interference the later material is to blame, in proactive interference the earlier material is. Students reliably recall that one runs forward and one runs backward but not which is which, because the direction is not recoverable from the word itself; fixing the culprit, not the victim, is the test that separates them.

Common Misconceptions

Forgetting happens because memories fade with time.

Time is not a cause; what fills the time is. Matched intervals produce far more forgetting when filled with similar learning than with rest or sleep, which is why interference, not decay, became the dominant account of everyday forgetting (McGeoch, 1932; Wixted, 2004).

Retroactive interference just means the new memory blocks the old one at the moment of recall.

Competition at retrieval is only one factor. When competition is removed by asking for both responses with unlimited time, first-list associations are still lost, showing genuine unlearning beyond momentary blocking (Barnes & Underwood, 1959).

An interfered memory is gone for good.

Much of the loss is a bias toward the most accessible route rather than destruction of the trace; controlled retrieval and the passage of time can restore first-list responses, so interference is often a failure of access rather than of storage (Lustig, Konkel, & Jacoby, 2004).

Glossary

A-B, A-C paradigm.

The standard interference design in which the same cues (A) are paired first with one set of responses (B) and then with another (C), maximizing competition for shared retrieval routes.

Consolidation.

The time-dependent stabilization of a newly encoded memory, during which the trace is fragile and vulnerable to disruption by new learning.

Decay theory.

The view that memories weaken automatically with the passage of time when not rehearsed; largely displaced by interference accounts for verbal forgetting.

Forgetting curve.

The declining function relating retention to the retention interval; interference theory attributes its shape to accumulating competition rather than to elapsed time.

Interpolated learning.

The activity placed between original learning and the retention test; its content, not the interval's length, determines the magnitude of retroactive interference.

MMFR (modified-modified free recall).

A test in which observers recall both responses to each cue with unlimited time, removing competition so that genuine unlearning can be measured.

Perseveration hypothesis.

Müller and Pilzecker's 1900 proposal that neural activity persists after learning to consolidate a memory, and that new activity interrupting it causes forgetting.

Proactive interference.

The complementary effect in which earlier learning impairs the retrieval of later material; the forward-acting mirror of retroactive interference.

Response competition.

The first factor of retroactive interference: newly learned responses vie with original responses at retrieval, producing overt intrusions.

Retrieval-induced forgetting.

The suppression of related memories caused by retrieving one of them, an active inhibitory process linking classic interference to executive control.

Retroactive interference.

The reduction in retention of earlier-learned material caused by learning that follows it; the interfering material is the newer material.

Spontaneous recovery.

The partial return of unlearned first-list associations over time after interpolated learning, paralleling the recovery of an extinguished conditioned response.

Unlearning.

The second factor of retroactive interference: genuine weakening of the original associations during interpolated learning, analogous to experimental extinction.

Wakeful rest.

A quiet, unfilled interval after learning that minimizes interpolated interference and reliably improves later retention of new material.

Key Researchers

Michael C. Anderson (b. 1962). Programme leader at the MRC Cognition and Brain Sciences Unit, University of Cambridge; he recast interference as active, executive-controlled inhibition and connected it to retrieval-induced forgetting. Google Scholar - ORCID - Faculty Page

John A. McGeoch (1897-1942). American functionalist psychologist; his critique of the law of disuse established that forgetting reflects interference among memories rather than the passage of time. Wikipedia - Wikidata

Arthur W. Melton (1906-1978). University of Michigan; with J. M. Irwin he formulated the two-factor account of retroactive inhibition and inferred the unlearning factor from the interference that outran overt competition. Wikipedia - Wikidata

Sergio Della Sala (b. 1955). Professor of Human Cognitive Neuroscience at the University of Edinburgh; with Dewar and Cowan he revived the minimal-interference account of forgetting, linking it to consolidation and amnesia. Google Scholar - ORCID - Wikipedia

Benton J. Underwood (1915-1994). Northwestern University; with Barnes he demonstrated unlearning through MMFR, and his broader programme established the reach of interference in verbal forgetting. Wikipedia - Wikidata

John T. Wixted (b. 1959). Distinguished Professor of Psychology at the University of California, San Diego; his synthesis recast interference in consolidation terms, arguing new learning disrupts the stabilization of older memories. Google Scholar - ORCID - Faculty Page

Frequently Asked Questions

What is retroactive interference in simple terms?
It is when learning something new makes it harder to recall something learned earlier, so the newer material is what disrupts the older memory (Melton & Irwin, 1940).

How is retroactive interference different from proactive interference?
The prefix names the culprit: in retroactive interference newer learning disrupts an older memory, whereas in proactive interference older learning disrupts a newer one (Underwood, 1957).

Why did retroactive interference undermine decay theory?
Because matched intervals produce much more forgetting when filled with similar learning than with rest, showing that the content of the interval, not its duration, drives the loss (McGeoch, 1932).

What are the two factors in the classic account?
Response competition, in which new responses intrude at retrieval, and unlearning, the genuine weakening of the original associations during interpolated learning (Barnes & Underwood, 1959).

When is retroactive interference strongest?
When the interpolated material is highly similar to the original and when more of it is learned, because shared cues with new responses compete most directly (McGeoch & McDonald, 1931).

Can new learning disrupt a memory before retrieval is even attempted?
Yes; a memory is fragile just after encoding, and new learning during that window can impair its consolidation, a loss unrelated to competition at test (Wixted, 2004).

Does resting after learning reduce forgetting?
A brief period of wakeful rest after learning preserves detail that an equivalent period of new activity erodes, by minimizing interpolated interference (Craig & Dewar, 2018).

Is an interfered memory lost permanently?
Often not; much of the deficit is reduced access rather than destruction, and controlled retrieval can recover apparently unlearned associations (Lustig, Konkel, & Jacoby, 2004).

References