During night-time driving hazardous objects often appear at mesopic light levels, which are typically measured using light meters with a spectral sensitivity that is only valid for photopic light levels. In order to develop suitable mesopic models a target detection experiment was performed in a driving simulator. While subjects drove along a winding road they had to respond to randomly presented circular targets at various eccentricities. The background luminance ranged between 0.01 and 10 cd/m², and was either white, yellow, red or blue. The RT, number of missed targets, and driving behaviour were measured. The results show that target detection and driving performance get poorer with decreasing background luminance and increasing eccentricity of the target, in particular for the red colour. At high driving speeds and low luminances subjects tend to neglect left off-axis targets. Luminance calculated with existing reaction-time-based mesopic models fits better to RT data than the widely-used photopic luminance. © 2006 The College of Optometrists Keywords: driving behaviour, driving simulator, mesopic vision, RT, traffic