Observational studies have suggested that psychometric psychosis liability and a functional polymorphism in the catechol-O-methyltransferase (COMT Val(158)Met) gene moderate the psychosis-inducing effect of cannabis. To replicate and extend this finding, a double-blind, placebo-controlled cross-over design was used in which patients with a psychotic disorder (n = 30), relatives of patients with a psychotic disorder (n = 12), and healthy controls (n = 32) were exposed to D-9-tetrahydrocannabinol (D-9-THC, the principal component of cannabis) or placebo, followed by cognitive assessment and assessment of current psychotic experiences. Previous expression of psychometric psychosis liability was also assessed. Models of current psychotic experiences and cognition were examined with multilevel random regression analyses to assess (i) main effects of genotype and condition, (ii) interactions between condition and genotype, and (iii) three-way interactions between condition, genotype, and psychometric psychosis liability. Carriers of the Val allele were most sensitive to D-9-THC-induced psychotic experiences, but this was conditional on prior evidence of psychometric psychosis liability. D-9-THC impacted negatively on cognitive measures. Carriers of the Val allele were also more sensitive to D-9-THC-induced memory and attention impairments compared to carriers of the Met allele. Experimental effects of D-9-THC on cognition and psychosis are moderated by COMT Val 158 Met genotype, but the effects may in part be conditional on the additional presence of pre-existing psychosis liability. The association between cannabis and psychosis may represent higher order gene-environment and gene-gene interactions.