Background: Negative affective processes may contribute to maintenance of paranoia in patients with psychosis, and vice versa. Successful treatment may break these pathological symptom networks. This study examined whether treatment with virtual reality based cognitive behavioral therapy (VR-CBT) for paranoia influences momentary affective states, and whether VR-CBT changes the adverse interplay between affective states and paranoia. Methods: Patients with a psychotic disorder (n = 91) were randomized to 16-session VR-CBT or treatment as usual (TAU). With the experience sampling method (structured diary technique) mental states were assessed for 6–10 days at baseline, posttreatment and 6-month follow-up. Multilevel analysis were performed to establish treatment effects and time-lagged associations between mental states, that were visualized with networks of mental states. Results: Average levels of paranoia (feeling suspicious [b = −032., p = .04], disliked [b = −49., p < .01] and hurt [b = −0.52, p < .01]) and negative affect (anxious [b = −0.37, p = .01], down [b = −0.33, p = .04] and insecure [b = −0.17, p = .03) improved more after VR-CBT than TAU, but positive affect did not. Baseline mental state networks had few significant connections, with most stable connections being autocorrelations of mental states. The interplay between affective states and paranoia did not change in response to treatment. A trend reduction in average intranode connections (autocorrelations) was found after VR-CBT (b = −0.07, p = .08), indicating that mental states reinforce themselves less after treatment. Conclusions: VR-CBT reduced paranoid symptoms and lowered levels of negative affect in daily life, but did not affect the extent to which mental states influenced each other. Findings do suggest that as a result of treatment mental states regain flexibility.