Twenty-three outpatients with schizophrenia (ICD-10 F20.xx) treated with clozapine (CZ) as monotherapy entered a prospective study on relapse prevention. Every 4 weeks, psychopathology was assessed by the Brief Psychiatric Rating Scale (BPRS), and plasma CZ and norclozapine levels were measured. Patients were enrolled after complete remission of positive symptoms for at least 4 months according to the psychosis cluster of the BPRS and at a mean of 3.3 years after their last hospitalization. At the time of enrollment, the median BPRS total score was 29 points (range, 19-48). Within 4 months, the baseline CZ plasma level was established as the mean of CZ levels from at least four subsequent measurements. These baseline plasma levels were considered as the optimal relapse-preventing plasma CZ levels in the individual patients. When the patients were enrolled, they were considered to be prone to relapse. Relapse was defined as clinical deterioration, hospitalization, or both. Plasma levels were considered a prognostic factor, and patients were defined as at increased risk if plasma levels decreased by more than 40% from baseline CZ plasma level. The effect of plasma CZ levels on clinical outcome was evaluated by a Cox regression with plasma level as a time-dependent covariate. Within 46 months of enrollment, 32 episodes of relapse events in 10 patients were available for evaluation. Seventeen patients had a plasma level decrease of more than 40% at some point. In 12 of these, the decrease was present for more than 12% of the observation period. Eight patients of this group relapsed, and three of these had to be rehospitalized. Two patients relapsed, although their plasma levels decreased by more than 40% for less than 12% of the observation period. Within the first 2 years, relapse-free survival curves illustrate that both groups (episodes under elevated risk and episodes not under elevated risk) had identical relapse patterns, but from then on the relapse risk increased rapidly in the group with longer exposure to elevated risk. In a Cox model with a 40% decrease of plasma CZ levels as a dichotomous time-varying explanatory covariate, the risk ratio is 6 (95% confidence interval = 2-19, p = 0.003). The 10 patients who relapsed exhibited safe plasma levels (less than a 40% decrease from their baseline levels) for only 210 months, and 13 nonrelapsing patients had plasma levels defined as safe for 426 months.