The basic objective of this study is to analyse the role of farm size and diversification in determining farmer’s total income from both farm and non-farm sources. Using data from NSS 70th Round Situation Assessment Survey, we estimate a log-linear regression model that relates farmer’s income to farm size, on-farm and off-farm diversification, and various other control factors representing farm, household and locational characteristics. We estimate this model for total income over the whole year and separately for the two seasons, kharif and rabi. We find that farm size has a negative relationship with farmer’s income per capita after controlling for various factors. Our results also lend support to our hypothesis that there is an optimal level of diversification that maximizes farmer’s income. We find that the optimal number of crops to be engaged in is 2 in both seasons, 2 animal husbandry activities in both seasons, 4 non-farm activities in kharif and 3 non-farm activities in rabi season. Comparing these estimates with the actual levels as per the NSS data shows that farmers on an average are already engaged in the optimal number of crops, but they are at sub-optimal level in terms of animal husbandry and non-farm activities. We also find that there is a minimum threshold level of education, viz., “Literate with Formal Schooling”, required to improve income levels. NSS data show that on an average, marginal and small farmers who constitute nearly 85 per cent of agricultural households have an education level below this threshold. Thus, improving their education levels is another point for policy intervention that can help raise their income levels. Another interesting finding from our analysis is that participating in MGNREGS may have an adverse impact on income levels, possibly via the opportunity cost of time spent in such public works.