As one of the measures to smooth income, this paper focuses on the diversification of labor allocation across activities. A key feature of this paper is that it pays particular attention to differences in the covariance between weather risk and agricultural wages and between weather risk and nonagricultural wages. We estimate a multivariate tobit model of labor allocation using household data from rural areas of Bihar and Uttar Pradesh, India. The regression results show that the share of the offfarm labor supply increases with the weather risk, and the increase is much larger in the case of nonagricultural wage work than in the case of agricultural wage work. Simulation results based on the regression estimates show that the sectoral difference is substantial, implying that empirical and theoretical studies on farmers' labor supply response to risk should distinguish between the types of off-farm work involved.