OBJECTIVE: The purpose of this study was to develop a model based on the authors' previous studies to identify which neuroleptic-treated schizophrenic patients are at risk of early relapse following drug withdrawal. METHOD: Clinical and CSF monoamine-related variables obtained for 50 male haloperidol-treated, schizophrenic patients were used in a logistic regression model to identify those who relapsed (N = 24) within 6 weeks after placebo substitution and those who did not (N = 26). RESULTS: The oral dose of haloperidol, weight, CSF norepinephrine, 3-methoxy-4-hydroxyphenylglycol and chromogranin A-like immunoreactivity, and the anxiety and paranoia subscale ratings of the Brief Psychiatric Rating Scale produced a model that correctly predicted 18 relapsers and 21 nonrelapsers. By including the interactions of paranoia subscale by CSF norepinephrine and anxiety by CSF norepinephrine, the model correctly identified 20 relapsers and 23 nonrelapsers with a sensitivity and specificity of 83% and 88%, respectively. CONCLUSIONS: Increased noradrenergic activity during chronic dopamine blockade may be an episode marker and may predict relapse within 6 weeks following haloperidol withdrawal in schizophrenia. Effective relapse prediction models have important practical implications for the treatment of schizophrenia and the understanding of the psychotic relapse process.