Complexity is a decisive feature of tax systems and a recent body of evidence reveals that it prompts taxpayers to underestimate taxes. Extending Mirrlees (1971) seminal optimal taxation model to allow for misperceptions, we investigate whether tax complexity may be a desirable policy tool in this environment. We think of tax complexity as the obfuscating features of the tax system and model it as an information cost. The higher the complexity, the more taxpayers prefer to rely on their priors about the tax scheme and the less elastic is their labor supply. We characterize the optimal policy -- namely the optimal tax schedule and the optimal tax complexity -- and show that complexity may be strictly positive at the optimum. This appears to be particularly relevant when priors are downward biased, a common result in most empirical studies.