Background <p>Punctate white matter injury on brain magnetic resonance imaging (MRI) is described in very preterm infants (&lt; 32&#xa0;weeks’ gestation) and is predictive of poorer developmental outcomes. The reliability of scoring and the incidence and evolution of white matter injury in moderate-late preterm infants is unknown.</p> Objective <p>To assess inter-observer variability in white matter injury using a published scoring system (UCSF system), and to describe changes over time in moderate-late preterm infants.</p> Materials and methods <p>Infants born between 32 + 0 and 36 + 6&#xa0;weeks’ gestation in the Auckland region underwent MRI scans as soon as clinically feasible after birth and again at term-equivalent age. De-identified scans were scored independently by two observers. White matter injury was graded as minimal (&lt; 3 lesions measuring &lt; 2&#xa0;mm), moderate (&gt; 3 lesions or lesions &gt; 2&#xa0;mm), or severe (&gt; 5% hemispheric involvement). Scores were compared between reviewers using weighted and unweighted kappa statistics interpreted using Cohen’s criteria. Incidences were compared between scans using generalised estimating equations.</p> Results <p>Scans of 101 infants were assessed. Inter-observer agreement was near perfect for the presence of white matter injury (<i>k</i> = 0.88 and 0.81 for the first and second scan respectively), and for the severity of white matter injury was near perfect at the first scan (<i>k</i> = 0.85) and substantial at the second scan (<i>k</i> = 0.80). The incidence of white matter injury detected by the two observers decreased between the first and second scans (30% to 22% and 29% to 19%), and severity also decreased.</p> Conclusions <p>This scoring system can be reliably applied in moderate-late preterm infants. White matter injury is common in moderate-late preterm infants but may be underestimated when MRI is performed close to term-equivalent age.</p> Graphical Abstract <p></p>