Aim: To identify factors that may predict and affect the risk of relapse in anti-N-methyl-D-aspartate receptor (NMDAR) encephalitis. Method: This was a retrospective study of an Italian cohort of patients with paediatric (≤18y) onset anti-NMDAR encephalitis. Results: Of the 62 children included (39 females; median age at onset 9y 10mo, range 1y 2mo–18y; onset between 2005 and 2018), 21 per cent relapsed (median two total events per relapsing patient, range 2–4). Time to first relapse was median 31.5 months (range 7–89mo). Severity at first relapse was lower than onset (median modified Rankin Scale [mRS] 3, range 2–4, vs median mRS 5, range 3–5; admission to intensive care unit: 0/10 vs 3/10). At the survival analysis, the risk of relapsing was significantly lower in patients who received three or more different immune therapies at first disease event (hazard ratio 0.208, 95% confidence interval 0.046–0.941; p=0.042). Neurological outcome at follow-up did not differ significantly between patients with relapsing and monophasic disease (mRS 0–1 in 39/49 vs 12/13; p=0.431), although follow-up duration was significantly longer in relapsing (median 84mo, range 14–137mo) than in monophasic patients (median 32mo, range 4–108mo; p=0.002). Interpretation: Relapses may occur in about one-fifth of children with anti-NMDAR encephalitis, are generally milder than at onset, and may span over a long period, although they do not seem to be associated with severity in the acute phase or with outcome at follow-up. Aggressive immune therapy at onset may reduce risk of relapse. What this paper adds: Relapses of anti-N-methyl-D-aspartate receptor encephalitis may span over a long period. Relapses were not associated with severity in the acute phase or outcome at follow-up. Aggressive immune therapy at onset appears to decrease risk of relapse.