Although antipsychotic maintenance treatment is widely recommended to prevent relapse in chronic psychoses, evidence-based guidelines do not provide clear indications on different maintenance treatment strategies, including continuing the antipsychotic at standard doses, reducing the dose, switching to another antipsychotic, or even stopping the antipsychotic. We aimed to compare the effectiveness of these maintenance treatment strategies, hypothesising the superiority of all strategies over stopping, and of continuing at standard doses over both switching and reducing the dose. We did a systematic review and network meta-analysis of randomized controlled trials (RCTs) that investigated antipsychotics for relapse prevention in adults with schizophrenia-spectrum disorders who were clinically stable, and which compared four treatment strategies: continuing the current antipsychotic at standard doses recommended for acute treatment; reducing the current antipsychotic dose; switching to a different antipsychotic; and stopping the antipsychotic and replacing it with placebo. We excluded RCTs with fewer than 25 individuals, a prerandomisation washout period greater than 4 weeks, a follow-up shorter than 6 weeks, and those recruiting treatment-resistant individuals. We searched MEDLINE, EMBASE, PsycINFO, CINAHL, CENTRAL, and online trial registers for published and unpublished RCTs from inception to Sept 1, 2021, combining terms describing all available antipsychotics, and terms describing continuation, maintenance, or long-term treatment for schizophrenia-spectrum disorders. Relative risks (RRs) and standardised mean differences were pooled using random-effects pairwise and network meta-analyses. We assessed risk of bias of each RCT with the Cochrane Risk-of-Bias 2 tool, and confidence of pooled estimates with CINeMA. The primary outcome was relapse prevention. The study protocol was registered in advance in the Open Science Forum registry. Of 3936 records identified, 119 records, reporting on 101 RCTs, were eligible, 98 of which (including 13 988 individuals) provided data that could be meta-analysed for at least one outcome. The mean proportion of female participants per study was 38% (range 0-100; median 39%, IQR 29-50), whereas for male participants it was 62% (range 0-100; median 61%, IQR 50-71), and the overall mean age was 38·8 years (range 23·2-63·9; median 39·3, IQR 35·0-43·9). Of the 98 RCTs meta-analysed, 89·8% were done in high-income and upper-middle-income countries. The ethnic group White or so-called Caucasian was the most represented (mean 56% participants per study), although this information was relatively scarce. All continuation strategies were significantly more effective in preventing relapse than stopping antipsychotic treatment, with a large risk reduction for continuing at standard doses (RR 0·37, 95% CI 0·32-0·43; number-needed-to-treat [NNT] 3·17, 95% CI 2·94-3·51) and antipsychotic switching (RR 0·44, 0·37-0·53; NNT 3·57, 3·17-4·25), and moderate risk reduction for dose reduction (RR 0·68, 0·51-0·90; NNT 6·25, 4·08-20·00). Continuing and switching antipsychotics did not differ significantly (RR 0·84, 0·69-1·02; with lower values favouring continuing), whereas reducing antipsychotic dose was outperformed by both continuing (RR 0·55, 0·42-0·71; NNT 4·44, 3·45-6·90) and switching (RR 0·65, 0·47-0·89; NNT 5·17, 3·77-18·18). Results were supported by moderate confidence of evidence and confirmed by secondary analyses and by several sensitivity and subgroup analyses, including removing studies with abrupt antipsychotic discontinuation or fast tapering (≤4 weeks). No tolerability differences emerged between treatment strategies. According to the Cochrane Risk-of-Bias tool, version 2, 16·8% of included RCTs had an overall high risk of bias for the primary outcome. We found moderate heterogeneity (τ2=0·13; I2=61%) and no overall incoherence for the primary analysis. Results were supported by moderate confidence of evidence and confirmed by secondary analyses. Contrary to our original hypothesis, we found that continuing antipsychotic treatment at standard doses or switching to a different antipsychotic are similarly effective treatment strategies, whereas reducing antipsychotic doses below standard doses is associated with higher risk of relapse than the other two maintenance treatment strategies and should therefore be limited to selected cases. Despite limitations, including moderate heterogeneity and moderate certainty of evidence, these results are of pragmatic relevance for clinicians, and should support the update of evidence-based guidelines. None.