While cognitive skills are known to play an important role in labour market success, empirical evidence is mainly concentrated in its effect on returns to schooling. Evidence on the role of cognitive skills in gender earnings gap decompositions is virtually absent. I use two approaches to investigate the potential for cognitive skills to affect the size and pattern of the unexplained component of the earnings gap (‘relative discrimination’) across the wage distribution, using data from the International Adult Literacy Survey (IALS). It has been shown that when the raw cognitive score is used to control for cognitive skills, the return to schooling cognitive skills is generally underestimated. Once a distinction is made about the origins of cognitive skills (acquired in school versus outside the school), the returns to cognitive skills can vary depending on their origin. I find that using the raw score to control for cognitive skills does not result in any significantly different estimates of the unexplained component of the gap compared to when cognitive skills are not controlled for. However, once cognitive skills by origin of skill are used in place of total cognitive skills, the results change substantially for three of the five countries examined.