Loss aversion is widely regarded as the most robust and ubiquitous finding in behavioural economics. According to the loss aversion hypothesis, the subjective value of losses exceeds the subjective value of equivalent gains. One common assumption in the literature is that this asymmetry represents a fundamental and stable feature of people’s preferences. In cumulative prospect theory, loss aversion is captured by the lambda (λ) parameter, which controls the steepness of the value function for losses. Estimates of λ by Tversky and Kahneman (1992) found evidence for considerable overweighting of losses in risky choice (λ = 2.25). But others find very different levels of loss aversion, with some reporting weak loss aversion or even loss neutrality. In order to assess what is the average level of λ reported in the literature, we set out to conduct a meta-analysis of studies in which the λ parameter of the cumulative prospect theory parameter was estimated from individual choices between risky prospects. We draw three conclusions. First, surprisingly few studies have estimated λ using risky choices, and there are only a few datasets suitable to perform model fitting. Second, much of the data are of poor quality, making it impossible to obtain precise estimates of the prospect theory’s parameters. Third, using a random-effect meta-analysis upon the available data, we found a small λ of 1.31, 95% CI [1.10, 1.53].