Abstract Individual crop insurance has been largely abandoned in developing countries and replaced by insurance pilots based on weather indices. These pilot schemes have encountered low demand. Research suggests that better-off farmers may already be insured via income diversification, their assets and social networks, and may achieve profit-maximising portfolios without formal insurance contracts. They would be interested in such contracts only if they reliably reduced their exposure to risk at lower costs than their self-insurance mechanisms. Conversely, poor farmers are not able to self-insure adequately, have to trade-off returns for reduced risk and could, therefore, benefit from a well-designed insurance. But they are cash/credit constrained and, therefore, cannot advance the money before sowing time to buy insurance that pays out only after the harvest. Index insurance, therefore, cannot be scaled up. Even if a few farmers purchase it, governments still will need to run relief programmes for the uninsured. Standard ways suggested to improve the index insurance, such as reducing basis risks, educating farmers and improving weather data, do not improve the ability of small farmers to purchase insurance and may not improve product design sufficiently to be competitive with self-insurance of the better-off farmers.