International Neuropsychiatric Disease Journal

Phantom limb pain is experienced in an absent body part and can persist long after the tissue and sensory receptors from which pain ostensibly originated have been removed. A neuropathic account correctly recognises the consequences of somatosensory injury, yet does not by itself explain the persistence, bodily localisation, episodic reactivation or variable response to peripheral and centrally directed treatments. This critical narrative review evaluates whether neurobiological memory provides a useful, empirically constrained framework for understanding phantom limb pain. Literature published from 1990 to 23 August 2026 was identified through targeted scholarly searching, citation tracing and verification of original clinical, experimental and synthesis papers. Evidence is examined across four non-equivalent meanings of memory: explicit recollection of pre-amputation pain, persistent sensorimotor representations, experience-dependent amplification of nociceptive circuits, and learned expectations about the missing body part. Reports of pain resembling earlier injuries support a phenomenological memory component in some individuals, but prospective findings and congenital phantom experiences refute the proposition that all phantom pain is recalled pain. Experimental interruption of peripheral input demonstrates that ongoing neural activity can remain necessary even when central representation has become reorganised. Conversely, neuroimaging reveals both cortical remapping and enduring hand representations, with recent longitudinal observations challenging a universal maladaptive-map account. Mirror-based, imagery, extended-reality, nerve-directed and neuromodulatory treatments offer complementary mechanistic tests, although small samples, imperfect controls and heterogeneous pain outcomes restrict causal inference. The most defensible interpretation is a distributed, dynamically maintained pain state in which peripheral drive, spinal gain, body representation and learned prediction can contribute in differing proportions. Memory is therefore better treated as a family of testable maintenance processes than as a singular stored pain trace. Future work requires prospective measurements before and after amputation, mechanism-specific perturbations, credible active comparators and validated within-person outcomes.