Smoking is strongly associated with schizophrenia. Although it has been widely assumed that this reflects self-medication, recent studies suggest that smoking may be a risk factor for schizophrenia. We performed two-sample bi-directional Mendelian randomization using summary level genomewide association data from the Tobacco And Genetics Consortium and Psychiatric Genomics Consortium. Variants associated with smoking initiation and schizophrenia were combined using an inverse-variance weighted fixed-effects approach. We found evidence consistent with a causal effect of smoking initiation on schizophrenia risk (OR 1.73, 95% CI 1.30–2.25, p < 0.001). However, after relaxing the p-value threshold to include variants from more than one gene and minimize the potential impact of pleiotropy, the association was attenuated (OR 1.03, 95% CI 0.97–1.09, p = 0.32). There was little evidence in support of a causal effect of schizophrenia on smoking initiation (OR 1.01, 95% CI 0.98–1.04, p = 0.32). MR Egger regression sensitivity analysis indicated no evidence for pleiotropy in the effect of schizophrenia on smoking initiation (intercept OR 1.01, 95% CI 0.99–1.02, p = 0.49). Our findings provide little evidence of a causal association between smoking initiation and schizophrenia, in either direction. However, we cannot rule out a causal effect of smoking on schizophrenia related to heavier, lifetime exposure, rather than initiation.