Twenty-two schizophrenic inpatients were treated for 3 weeks with three randomly fixed oral doses of haloperidol (10, 20, or 30 mg). Analysis of the results by a nonlinear regression model revealed a curvilinear relationship between haloperidol levels in plasma and clinical response, as assessed on the Brief Psychiatric Rating Scale (pseudo-R2 = 0.85, F = 17.7, p < 0.001, correlation between coefficients ranged from 0.99 to -0.52). This curve defines roughly three drug level ranges (low, < 5.5 ng/ml; optimal, 5.5 to 14.4 ng/ml; and high or toxic, > 14.4 ng/ml), which are significant for clinical practice. Patients with high levels improve to a lesser extent or even worsen in negative symptoms, showing a nonstatistically significant trend to present more extrapyramidal symptoms. Our data thus support the existence of a therapeutic window for haloperidol. Schizophrenic patients with acute exacerbation and drug levels in this range would have a greater probability of global clinical improvement.