Can Your Attachment Style Change? What the Research Actually Shows

People ask whether attachment style can change usually after they have read enough to recognise their own pattern and feel a quiet dread that the description is also a sentence. It is a reasonable fear and, fortunately, not an accurate one.

Attachment style can change. It is a learned set of expectations about whether closeness is reliably safe, formed through repeated early relational experience, and learned patterns are revisable through later experience that contradicts them. Research on what is called earned security documents people who developed insecure patterns early and moved toward secure functioning as adults. The change is real, but it is typically slow, uneven, and easier to see in hindsight than in progress.

---

Key Takeaways

• Attachment style is a learned pattern of expectations, not a fixed personality trait.
• Earned security describes adults with difficult early histories who developed secure functioning later, the clearest evidence that change is possible.
• Change usually happens through sustained relationships that contradict the original expectation, through therapy, or both.
• The process is slow and non-linear; progress shows up as the old reaction resolving faster, not disappearing.
• Patterns can also shift in the insecure direction under stress or betrayal, and are revisable from there too.
• Attachment has a stable general component and a relationship-specific component that responds to the particular partner.

---

Is attachment style fixed or learned?

The premise of attachment theory is that early relational experience builds internal working models: implicit expectations about whether closeness can be relied on and what one has to do to keep it. Bowlby's original framing (Bowlby, 1969) treated these models as formed early but open to revision in light of later experience, not as permanent imprints. The pattern is learned, which means it carries the property every learned pattern carries: it can be relearned.

This matters because the language of attachment "styles" can imply fixed categories, like blood type. The research does not support that reading. Mikulincer and Shaver's synthesis of decades of adult attachment work (Mikulincer & Shaver, 2007) describes attachment security as varying both across people and within the same person over time and across relationships. The categories are useful shorthand for a tendency, not a permanent classification. The foundational frame is laid out in what is attachment theory.

What does the research on earned security actually show?

The strongest evidence that the early pattern is not a ceiling comes from work on earned security. Using narrative interview methods, attachment researchers found a group of adults who had clearly difficult early caregiving histories but who could nonetheless reflect on those histories coherently and relate to current partners and children in a secure way. They had not had secure attachment; they had developed it.

The mechanism behind earned security is usually some combination of a sustained corrective relationship and the development of a coherent narrative about the early history, often through therapy. The relationship provides repeated experience that contradicts the original expectation. The narrative work converts an unprocessed history into one the person can hold without it driving present behaviour. Neither alone is reliably sufficient; together they account for most documented movement toward security. The destination this points toward is described in the secure attachment guide.

The coherence finding is worth dwelling on because it is counterintuitive and practically important. In the narrative interview research, what distinguished earned-secure adults was not that they had had better childhoods, by definition they had not, but the way they could now talk about them: with detail, without idealising or dismissing, and without being flooded by the material as if it were still happening. The early history had been metabolised into a story rather than left as a live wire. This implies something specific about the work: the target is not erasing the past or reframing it as fine, but reaching a state where it can be described accurately and held steadily. That is a more achievable and more honest aim than the "heal it so it no longer hurts" framing self-help often implies.

It also clarifies why some people with objectively gentler histories remain insecure while some with harsher ones do not. The predictor is not the severity of the input but the degree to which it has been integrated. This is not a claim that severity is irrelevant; it plainly raises the difficulty. It is a claim that integration, not the absence of harm, is the active ingredient, which is why the route runs through coherent processing and corrective experience rather than through having been lucky.

Why is the change so slow and non-linear?

The pattern was not built by an event; it was built by repetition. The expectation that closeness is unreliable, or that needs must be handled alone, was confirmed enough times early on that it became the default reading of ambiguous relational situations. Defaults built by repetition tend to revise by repetition, which is inherently slow.

Non-linearity comes from a specific feature of the pattern: it reasserts under stress. A person can relate securely for months and then, under a threat that resembles the original one, drop back into the old reaction with full intensity. This is not failure or regression to baseline; it is the older model still being available and getting selected when the system is taxed. Progress is better measured by how the reaction resolves than by whether it still occurs. The reaction arriving less often, feeling less total, and being followed by repair rather than escalation is the actual evidence of change. The pursuit-withdrawal version of this is detailed in the anxious-avoidant trap.

What actually moves the pattern?

Three things account for most documented movement, and they tend to work together rather than independently.

The first is a sustained relationship that consistently contradicts the original expectation. For an anxiously attached person, that is a partner whose availability is reliable enough that vigilance slowly stops being necessary. For an avoidantly attached person, it is a partner who does not punish closeness, so distance slowly stops being the only safe position. The corrective has to be repeated and reliable, because a long history is not outweighed by a few good experiences.

The second is reflective work that builds a coherent account of the early history. This is much of what attachment-informed therapy does. The point is not to relive the history but to make it narratable, because an unprocessed history tends to operate as present-tense reaction rather than as past-tense memory.

The third is attention to the pattern in real time: noticing the bodily signature of the old reaction early enough to respond to it differently rather than from it. This is the slowest-building skill and the one most within the person's own control. The interaction of these patterns across different pairings is mapped in attachment style pairs.

There is also a small but encouraging body of work suggesting attachment can be a deliberate target, not only a by-product of circumstance. Research on volitional change in attachment (Hudson, Fraley, & colleagues, 2020) found that people who wanted to become more secure could shift measurably over months when the wanting was paired with concrete changes in how they actually behaved in close relationships, mirroring the volitional-change findings in trait psychology. The effect was modest and gradual rather than transformative, which is consistent with everything else known about this pattern: intention sets a direction, repeated contradicting behaviour does the moving, and time does the consolidating. What this rules out is the fatalistic reading, common in popular attachment content, that you simply are your style and the most you can do is warn partners about it. The evidence does not support that ceiling.

It is also worth saying what does not reliably move the pattern, because the absence is informative. Reading about attachment, taking quizzes, and being able to name your style with precision are useful for orientation but are not, by themselves, change. Insight tends to alter how a person narrates the pattern long before it alters the pattern itself, which can produce a frustrating phase of understanding exactly what you are doing while still doing it. That phase is not failure; it is the expected gap between the conscious account and the somatic default, and it closes through experience rather than through more reading.

Can attachment style get worse?

Yes, and saying so is part of taking the model seriously. A securely functioning person can shift toward insecurity under sustained relational stress, betrayal, or loss, particularly when the experience confirms a fear the person had previously outgrown. The model updates on current evidence, not only early evidence.

This is not usually a permanent reversal. It is the same revisability that allows movement toward security, operating in the other direction. A pattern that became more insecure because experience confirmed an old fear tends to be revisable through experience that contradicts the confirmation, by the same slow route. The symmetry is actually reassuring: it means the pattern is responsive rather than locked, in both directions. How this interacts with stable personality tendencies is explored in personality compatibility in relationships.

What does realistic change look like day to day?

Realistic change rarely feels like becoming a different person. It feels like the same reaction with a different course. The anxious spike still arrives, but it is recognised sooner, is less convincing, and resolves through a steadier action rather than an escalating one. The avoidant pull to withdraw still arrives, but it is noticed as a regulation strategy rather than experienced as a verdict, and it is followed by a return rather than a longer absence.

People often miss this because they are looking for the pattern to be gone and interpret its continued presence as evidence that change is not happening. The more accurate reading is that the pattern's persistence with a changed trajectory is exactly what successful change looks like. Disappearance is rarely the realistic target; a different relationship to the same reaction is.

This reframing also changes what counts as a setback. Under the disappearance model, a strong recurrence of the old pattern, a spike of pursuit, a stretch of withdrawal, reads as proof the change was illusory, which tends to trigger the very pattern it describes. Under the trajectory model, a recurrence is expected data, most informative for how it resolves rather than whether it occurred. People who hold the trajectory model tend to recover from relapses faster, partly because they are not adding a second layer of alarm about the relapse on top of the relapse itself. The belief about how change works is, in this narrow but real sense, part of the mechanism of change.

---

The early pattern is a strong prior, not a permanent identity. It was learned, which is precisely why it can be relearned, and the evidence that people do relearn it, slowly and unevenly, is some of the more hopeful findings in relationship psychology. The work is mostly the accumulation of contradicting experience and the attention to notice it landing.

Take the InnerPersona assessment — see your current attachment pattern alongside the trait and values dimensions that shape how readily it tends to revise, so you know what you are actually working with.

Read next: The secure attachment guide