This paper investigates the relationship between idiosyncratic risk and returns for individual securities within a generalized autoregressive conditional heteroskedascticity (GARCH)-in-mean framework. We demonstrate that, on average, 15% of stocks exhibit a significant relationship between returns and risk, of which 9% are positive. These proportions vary over time and with model specifications. Some characteristics influence the probability of a positive and a negative relationship, while others appear to affect only one, but not the other. This evidence implies that the factors that explain a positive connection between idiosyncratic risk and returns are different from the factors that explain a negative connection.