Until late 1980s, the rural households in the developing countries were completely dependent on the farm sector for their livelihood. They hardly resorted to alternative sources of income like the Rural Non-farm Sector (hereinafter, RNFS). However, in late 1980s, amidst the occurrence of the structural transformation, the income portfolios of the rural households witnessed a shift towards the RNFS and RNFS diversification became a norm across these households. Nevertheless, the vitality of the farm sector cannot be overlooked because the sector not only forms a major contributor in terms of providing employment to the rural population, but also acts a major source of food security, nutrition, poverty alleviation and overall growth and development of a country. Howbeit, the farm production is unequivocally affected by diversification into the RNFS via micro-economic consequences on the behaviour of the farm households. Once diversified into RNFS, farm production depends on whether or not income from the RNFS is invested in farm activities. This investment decision may in turn depend on farmers’ attitude to the risk they face in agricultural production due to various climatic and non-climatic factors. Against this backdrop, we make an attempt to develop a theoretical model to show that (i) farmers’ decision to invest RNFS income on farm activities varies with risk aversion of the households (ii) the impact of RNFS income on farm sector production is ambiguous. Underpinned by the theoretical model, we carry out the empirical estimation of the effect of RNFS diversification on farm sector performance (measured by farm productivity, here) using Wooldridge’s IV-2SLS Method and Vella-Verbeek’s Model. Both the methods control for the endogeneity of the RNFS income as well as for the plausibility of the sample selection bias of the households. The empirical estimation is supported by the VDSA micro-level dataset, wherein eight Indian states are studied for a panel of five years (2010-14). Our empirical analysis not only renders attention on the impact of RNFS income on the performance measure (farm productivity), but also on the channel (investment in farm inputs) through which farm performance may improve. The empirical results show that the impact of RNFS income on farm productivity is non-linear in nature, with farm productivity first falling and then rising with the level of RNFS income. Similar results are unveiled while assessing the impact of RNFS income on farm input expenditure. We also find that that the farm investment/farm input expenditure varies with the risk-averseness (estimated using Tadeo and Wall, 2011) of the households.