IMPORTANCE Cannabis use after first-episode psychosis is associated with poor outcomes, but the causal nature of this association is unclear. OBJECTIVE To examine the precise nature of the association between continued cannabis use after the onset of psychosis and risk of relapse of psychosis. DESIGN, SETTING, AND PARTICIPANTS This prospective cohort study followed up for at least 2 years after the onset of psychosis 220 patients who presented to psychiatric services in South London, England, from April 12, 2002, to July 26, 2013, with first-episode psychosis. Longitudinal modeling (fixed-effects analysis, cross-lagged path analysis) was used to examine whether the association between changes in cannabis use and risk of relapse over time is the result of shared vulnerability between psychosis and cannabis use, psychosis increasing the risk of cannabis use (reverse causation), or a causal effect of cannabis use on psychosis relapse. INTERVENTIONS Exposure to cannabis within the first and second years after onset of psychosis. MAIN OUTCOMES AND MEASURES The main outcome measure was relapse of psychosis, defined as subsequent hospitalization for psychosis. Effect of cannabis use status in the first year (Ct1) and second year (Ct2) and pattern of cannabis use continuation in the first year and second year were modeled for risk of relapse in the first year (Rt1) and risk of relapse in the second year (Rt2) after psychosis onset. RESULTS A total of 220 patients with first-episode psychosis were included in the analysis (mean [SD] age, 28.62 [8.58] years; age range, 18-65 years; 90 women [40.9%] and 130 men [59.1%]). Fixed-effects models that adjusted for time-variant (other illicit drug use, antipsychotic medication adherence) and time-invariant (eg, genetic or premorbid environment) unobserved confounders revealed that there was an increase in the odds of experiencing a relapse of psychosis during periods of cannabis use relative to periods of no use (odds ratio, 1.13; 95%CI, 1.03-1.24). Change in the pattern of continuation significantly increased the risk (odds ratio, 1.07; 95%CI, 1.02-1.13), suggesting a dose-dependent association. Cross-lagged analysis confirmed that this association reflected an effect of cannabis use on subsequent risk of relapse (Ct1→Rt2: β = 0.44, P = .04) rather than an effect of relapse on subsequent cannabis use (Rt1→Ct2: β = −0.29, P = .59). CONCLUSIONS AND RELEVANCE These results reveal a dose-dependent association between change in cannabis use and relapse of psychosis that is unlikely to be a result of self-medication or genetic and environmental confounding.