We study decision-making of judges in an experimental settingresembling real world judicial decision making. We gave to 312 future judges 48 vignettes built from real datarelated to divorce cases involving children. We compare two different subject pools : judges who were asked to set child support awards with a guideline and judges who were asked to set child support awards without any guideline. We found that the introduction of a guideline contributes to reduce the disparity between judges (i.e. the variance for like cases is lower when the subjects have the opportunity to use the guideline) but this effect is not systematic, an increase in heterogeneity being observed for some specific cases.