Economic liberalisation policies introduced since the early 1990s helped in accelerating growth of the economy, but it also increased farmers’ distress especially among marginal and small farmers in the backward states. The accelerated farm mechanisation, increased share of purchased inputs, fluctuation in prices and higher wage rates increased the vulnerability of small farmers than large farmers. In this context, historical inverse relationship between farm size and productivity has changed into positive one, with larger farms getting the advantage of large scale mechanisation and technology. The paper revisits the farm size and productivity debate at the state level by using unit level data of cost of cultivation for rice crop. The paper has found that, there is convergence of yield of rice across states mainly helped by convergence in fertiliser, irrigation and farm machinery use. However, there was widening gap between bottom 25 per cent and top 25 per cent (based on farm size) of the farmers in terms of yields, gross returns, profitability. The condition of tenant-small farmers was more precarious due to high land rents (50 per cent of total cost). The share of loss making farms is 17 per cent among bottom 25 per cent of the farmers compared to only 3 per cent among top 25 per cent of the farmers. The distress of small farms is aggravated by higher risk in profitability. Most of the farmers in states like Punjab, Haryana, Andhra Pradesh, Tamil Nadu and Gujarat are getting reasonable profits, while farmers of Maharashtra, Jharkhand, Assam, Bihar and Orissa are earning meagre profits. Use of modern inputs (farm machinery use and fertiliser use) helped in increasing profitability, while use of traditional inputs (animal labour and manure) are associated with losses. There was a convergence in the use of modern inputs across states, which have the potential to reduce inter-state disparities in profits and yields. The regression results show that there was a positive relation between farm size and profitability after controlling for state structural variables and input use.