Background Smoking may have a beneficial effect on either schizophrenic symptoms or antipsychotic side-effects, but studies are hampered by the lack of control of confounding factors. Aims To explore the self-medication hypothesis in a large sample of stable out-patients with schizophrenia. Method Symptoms, assessed with the Positive and Negative Syndrome Scale (PANSS), and number of hospitalisations were compared in 250 out-patients with DSM–IV schizophrenia classified into three categories: highly dependent smokers, mildly dependent smokers and non-smokers. Log-linear analysis was used to control for potential confounding and interacting variables. Results High PANSS total scores and positive symptoms were less frequent in mildly dependent smokers than in non-smokers or highly dependent smokers. The highly dependent smokers had the worst outcome. Conclusions The data do not generally support the self-medication hypothesis but rather suggest a complex interaction between nicotine dependence and schizophrenic symptoms.