We derive testable conditions ensuring that the income tax is optimal when agents are ex-ante identical but face idiosyncratic income risk. The optimal tax depends positively on both absolute risk aversion and risk variance and negatively on labor supply elasticity and absolute prudence. The comparison with the formula of the optimal non-linear income tax provides the restrictions on both the preferences and the income distribution conditional on effort ensuring that the optimal tax is indeed linear. In general it requires that the ratio of absolute prudence to absolute risk aversion be no less than two; if the income density has a linear likelihood ratio, it requires a (generalized) logarithmic consumption utility. Under HARA utility and linear or logarithmic likelihood ratios, explicit solutions for the optimal non-linear income tax are derived.